(navigation image)
Home American Libraries | Canadian Libraries | Universal Library | Community Texts | Project Gutenberg | Children's Library | Biodiversity Heritage Library | Additional Collections
Search: Advanced Search
Anonymous User (login or join us)
See other formats

Full text of "Exploring University Mathematics 1"

University Mathematics 1 


Bedford College, London 

The Commonwealth and Inter- 
national Library of Science 
Technology Engineering and 
Liberal Studies 










510, I 


Joint Chairmen of the Honorary Editorial Advisory Board 



Publisher: ROBERT MAXWELL, M.C., M.P. 

General Editors: W. J. LANGFORD. E. A. MAXWELL 







Edited by 



E»fe7 \^o5g- 


Pergamon Press Ltd., Headington Hill Hall, Oxford 
4 & 5 Fitzroy Square, London W.l 

Pergamon Press (Scotland) Ltd., 2 & 3 Teviot Place, Edinburgh 1 

Pergamon Press Inc., 44-01 21st Street, Long Island City, 
New York 11101 

Pergamon of Canada Ltd., 6 Adelaide Street East, Toronto, Ontario 

Pergamon Press (Aust.) Pty. Ltd., 20-22 Margaret Street, Sydney, 
New South Wales 

Pergamon Press S.A.R.L., 24 rue des Ecoles, Paris 5 e 

Vieweg & Sohn GmbH, Burgplatz 1, Braunschweig 

Copyright © 1967 Pergamon Press Ltd. 
tion 1967 

1 3 NOV 198fe 

This book is sold subject to the condition 

that it shall not, by way of trade, be lent, 

resold, hired out, or otherwise disposed 

of without the publisher's consent, 

in any form of binding or cover 

other than that in which 

it is published. 

(2852/67) • 




G. T. Kneebone, M. Sc., Ph. D. 
Reader in Foundations of Mathematics 
at Bedford College, London. 


J. R. Ellis, Ph.D. 

Assistant Lecturer in Applied Mathematics 
at Bedford College. London. 


S.J.Taylor, Ph.D. 
Professor of Pure Mathematics 
at Westfield College, London. 

4 WAVES 44 

P. Chadwick. Ph. D. 

Professor of Mathematics in the University of 
East Anglia at Norwich. (Formerly Senior Lecturer 
in Applied Mathematics at Sheffield University.) 


J.H.E.Cohn, M.A., D.Phil. 
Lecturer in Mathematics 

at Bedford College, London. 




M. Levison, Ph.D. 

Lecturer and Acting Head of Department of 
Computer Science, Birkbeck College, London. 


H. G. Eggleston, M. A., Sc. D. 
Professor of Pure Mathematics 
at Bedford College, London. 




By Professor H. G. Eggleston 

Head of the Mathematics Department, Bedford College, 


Tun lectures in this book were given at the 1965 Bedford 
College Easter Conference in Mathematics. These confer- 
ences have been held annually since 1963 when we initiated 
the programme with what we believe to have been the first 
conference of its kind held by any university in Britain. 
Our aims were (a) to increase the contact between schools and 
universities, (b) to enable pupils from different schools to 
meet and discuss topics of common interest and (c) to 
stimulate interest by introducing students in their last or 
penultimate years at school to branches of mathematics which 
would be novel to them. The lectures are primarily designed 
for students about to embark upon a degree course of which 
mathematics is a major part. As well as students, teachers 
of mathematics have also taken part in the conferences. 
Although those attending are drawn from schools in all parts 
of the country the number involved each year is, unfortu- 
nately, very limited. This year the organizers of the confer- 
ences felt that these lectures, given by professional mathe- 
maticians on subjects of current mathematical interest and 
yet assuming little mathematical background, would be of 
interest to a wider public. It was therefore decided to 
publish them in a book and so increase the "audience" many 



The seven lectures comprising the chapters of this book 
formed the programme for the 1965 Easter Conference in 
Mathematics at Bedford College, London. Those attending 
the conference were mostly in their last year at school and 
intending to read for a degree in mathematics at university 
the following year, or teachers concerned with this level of 
work in the schools. The scope of the lectures is fairly wide 
and is divided between pure mathematics and applied 
mathematics, with a natural bias towards the former at 
this level. Each lecture is quite independent, so that getting 
"lost" in one lecture does not mean that a subsequent lecture 
is unintelligible. (This, of course, is less important in the 
book, as the reader has time to take each lecture as 
slowly as necessary for complete comprehension.) Wherever 
possible, a list of suggestions for further reading is given 
at the end of the chapter. 

Four of the lectures were given by members of the Mathe- 
matics Department of Bedford College, whilst for the other 
three we were very pleased to welcome staff from other 
colleges or universities who were interested in taking part 
in the conference. Each lecturer chose a subject in which 
In- is an expert, either as a teacher or as a research worker. 

As far as possible in each conference, at least one lecture 
is given on the lecturer's research work, although the topics 
which may be presented at this level are rather limited. 
This year Dr. Cohn talked about a paper which he had 
published as recently as 1963. Although the problem on 
Square Fibonacci Numbers is simple to state and can be 



solved by elementary methods, the solution has eluded able 
mathematicians for many years and was only discovered by 
the exercise of considerable ingenuity. 

Professor Eggleston's lecture on the Isoperimetric Problem 
was given to an invited audience of the more advanced pupils 
(i.e. those who had already passed A-level), the school 
teachers and university lecturers. The mathematical level 
of this lecture is slightly higher than that of the other six 
which comprised the main programme for the conference, but 
taken at the reader's own pace it should be comprehensible 
and of considerable interest to all readers. 

I should like to thank all those who have taken part in the 
writing and proofreading of the book, and the Pergamon 
Press for the care which they have given to the production 

of the book. 

N. J. Hardiman 

Bedford College, London 
June, 1965 


Sets and Functions 

G. T. Kneebone 

As everyone knows who has studied the calculus, one of 
the most important concepts in mathematics is that of 
functional dependence. In this lecture I propose to glance 
at the general notion of function in order to see how it has 
been reinterpreted as a result of investigations made in recent 
times into the foundations of mathematics. 

Up to about 50 years ago it was taken generally for granted 
that the business of the mathematician is simply to do 
mathematics, that is to say either to use his expert knowledge 
in applying mathematical techniques to problems of science 
or engineering or else to extend mathematical knowledge 
itself by developing entirely new theories or adding further 
details to theories already in existence. From time to time 
a philosopher or philosophically-minded mathematician might 
crop up, who would ask questions concerning what mathe- 
matics is about or what constitutes valid mathematical proof — 
but such persons were thought of as being very much on the 
fringe of the subject. But today all this is radically changed 
and foundational studies are accepted, and even welcomed, 
as belonging to the main body of mathematics. Indeed, some 
of the most spectacular advances in modern mathematical 
research are due as much to the efforts of mathematical 
logicians as to those of mathematicians of the more traditional 


Such a state of affairs is not wholly new, for at certain 
crucial moments in the history of mathematics accepted 
mathematical ideas have turned out to be less transparent 
than had hitherto been supposed. It will be sufficient to 
recall two such episodes. 

(1) In ancient Greece, whenever mathematicians dealt with 
magnitudes they envisaged these not as abstract numbers but 
rather as lengths or areas; and the earlier Greek geometers 
thought it obvious that any line which can be obtained from 
a given line by simple geometrical construction can be 
measured exactly as a multiple of a sufficiently small sub- 
multiple of that line. But eventually this belief was dispelled 
by Pythagoras, who demonstrated that the diagonal of a 
square is not commensurable with the side (i. e. that the 
square root of 2 is an irrational number). This discovery 
of Pythagoras revolutionized the whole conception of magni- 
tude and led to the creation of the famous theory of 
proportion contained in Book V of Euclid's Elements, and 
also in a later age, much nearer our own day, to Dedekind's 
theory of real numbers. 

(2) The subject of geometry was conceived by Euclid and 
his contemporaries as a study of the properties of space as 
it really is — properties that are crudely exemplified in our 
physical constructions and measurements — and geometry 
was accordingly presented as a deductive theory based upon 
a small number of principles that were accepted as indubi- 
table. Such a view of the nature of geometrical knowledge 
endured for more than 2000 years; but it had, nevertheless, 
to be discarded when non-euclidean geometries were found 
to be a theoretical possibility, and Euclid's set of axioms 
became only one system among many possible ones. From 
that time on, euclidean geometry has been interpreted as a 
hypothetical theory instead of a factual theory of objective 
reality — and the axiomatic method, as thus modified, today 
dominates the whole of pure mathematics. 


Discoveries of a foundational character can thus have 
far-reaching repercussions within mathematics itself; and 
we shall now see how this has happened with the concept 
of function. To reach adequate understanding of a mathe- 
matical concept or a mathematical theory, of any but the 
most abstract kind, one must learn something of its history 
and something of its logic. On the whole we would expect 
the logic to count for more than the history — from a 
strictly mathematical point of view of course, though not 
necessarily from the rather different point of view of philo- 
sophy. In the present instance I shall begin with history, 
since the concept of function is not so much a single abstract 
idea as an orientation of mind that has found different 
expression in different ages. 

For the Greeks, the idea of function did not yet exist. The 
Greek theory of magnitudes (as presented by Euclid, for 
example) is essentially a static theory, concerned with 
discrete quantities and fixed relations between them, whereas 
the concept of a function arose in the first instance out of 
the consideration of quantities which are susceptible of 
variation in accordance with some fixed law. The mathe- 
matical concept in fact came into being in the seventeenth 
century, in close association with a radically new frame of 
thought in natural science. Whereas ancient and mediaeval 
science had been qualitative the new science was quantitative, 
and the aim of scientists was now to discover mathematical 
laws which characterize the interdependence of the physical 
quantities in terms of which the relevant state of a given 
system can be described. In some simple cases the relation- 
ship in question can be expressed by an algebraic equation 
(as in the case of Boyle's law pv = k) but more usually a 
differential equation is needed. In either case, however, 
the physical law is identified with a functional relationship. 

As long as we are thinking of dependencies between 
physical quantities, the ideas of variable and function seem 
very natural. The physical magnitude, e.g. the pressure of 


a gas, is something which can be directly measured whenever 
required, and which then has a well-defined value. And one 
quantity may be considered as a function of another when 
the value of the first is uniquely determined by the value 
of the second. But when we translate these considerations 
into the language of pure mathematics we no longer have 
any observable or measurable quantities to fall back upon, 
and our "variables" are then mere phantoms. Indeed, one can 
argue that the very notion of a variable quantity, conceived 
abstractly, is a contradiction in terms. It was in fact argued 
long ago by Zeno of Elea, in one of his famous paradoxes, 
that a flying arrow can never be moving, since at any instant 
it occupies precisely the space in which it is situated. 

For a long time mathematicians managed to evade the 
difficulty concerning the nature of abstract variables by 
relying upon geometrical intuition; and since they were 
already very much at home with geometrical relationships 
they were able to develop a detailed theory of functions 
by such means. Functions were thought to be represented 
adequately by their graphs; and a function was said to 
be continuous, for example, if its graph was free from 
breaks. On this basis a simple proof could be given of the 
fundamental theorem that if a continuous function / is 
positive for x = a and negative for x = b then there is a 
number c between a and b such that f (c) = 0. For we 
cannot pass from above the ^r-axis to below the #-axis by a 
continuous path without crossing the axis at least once. 

Gradually, however, the more careful mathematicians came 
to distrust this kind of semi-intuitive reasoning; and Cauchy 
and Gauss, in particular, tried to reformulate analysis (i. e. 
the theory of functions) strictly in terms of abstract numbers. 
While almost all the familiar theorems continued to be 
accepted as valid, this was only because ways were now 
found of proving them in a rigorous manner. 

Criticism of the intuitive approach to mathematics was 
pressed even further by Dedekind and Peano, both of 


whom were pioneers in raising the critical study of the 
foundations of mathematics to the dignity of an independent 
and essential discipline. One of Dedekind's decisive acts 
was his uncompromising rejection of the intuitive notion of 
continuity as an adequate basis for a theory of continuous 
functions; and it is mainly due to this initiative that the sort 
of plausibility argument that is exemplified by our earlier 
"proof" that any continuous function which changes sign 
must attain the value zero is no longer acceptable as a 
mathematical demonstration. 

It was not Dedekind, however, but Peano who eventually 
dealt the death-blow to intuitive analysis by giving an 
analytical definition of a space-filling curve. According to 
the traditional view every function has a graph, and every 
pair of equations x = f (t), y = g (t), in which / and g are 
given functions and t is restricted to some given range, 
defines an arc of a curve. Peano devised a pair of functions 
f and g with the property that, for any xo such that < Xq < 1 
and any y such that < y <, 1, there is a t such that 
0<t ^l, x = f (to), and y = g (to). In other words he 
defined parametrically a "curve" which passes through every 
point of the square region bounded by the four straight lines 
x = Q, x = I, y = 0, and y — 1. Since a curve in the intuitive 
sense is a locus with length but no breadth, it cannot fill up 
an area; and there is thus a fundamental incompatibility 
between the intuitive notion of curve and the more formal 
notion of the locus determined by a parametric representation. 
The moral of this, as Peano saw it and as the entire 
mathematical world came by degrees to see it also, is that if 
mathematical theorems make statements about functions in 
an abstract sense they cannot be proved by appeal to what 
is "obviously" the case for curves in a totally different 
intuitive sense. 

Since 1890, the year of publication of the paper in which 
Peano defined his anomalous curve, mathematicians have 
systematically purged the theory of functions of every vestige 



of reliance on geometrical ideas; and the theory in the form 
in which it is normally presented today is wholly abstract. 
But since, even in the most formal mathematics, we cannot 
produce something out of nothing, we must begin with 
something that is admitted as logically prior to mathematics, 
even though it can no longer be supplied by spatial intuition. 
There is now fairly general agreement that what we must 
presuppose consists of (1) the basic principles of logic, which 
are implicit in the very conception of deductive reasoning, 
and (2) certain very general notions such as that of set, 
which are themselves closely allied to abstract logic. Using 
only such materials, which lie at the furthest extreme of 
generality and abstractness, we can in fact refashion not only 
the theory of functions but the whole of pure mathematics. 

A function is essentially a correspondence, whereby a 
value jjq of the dependent variable y is associated with any 
arbitrarily chosen value xq of the independent variable x, 
and it is thus uniquely determined by the totality of all such 
associated pairs of numbers {xo, yo). If we denote by R the 
totality or set of all (real) numbers, and by R X R the set of 
all ordered pairs (x, y) of two real numbers x and y, every 
function / gives rise to a certain subset F of R X R, namely 
the set of all ordered pairs (x, y) such that y = / (x). And 
conversely, every subset F of R X R with the property that, 
for every number x, there is precisely one number y such 
that the ordered pair (x, y) belongs to F, determines a 
function /. The concept of function is thus definable in 
set-theoretic terms; and the same is true in principle of 
every other mathematical concept — though it may well be 
that the set-theoretic definition of what appears to be a 
simple intuitive concept is formidably elaborate* 

In discussing the concept of function we have taken the 
concept of number for granted, but naturally this concept 
also must be made independent of intuition by abstract 
set-theoretic definition. How this is done is, in fact, another 
story, and by no means a short one. I have already referred 


to Dedekind's theory of real numbers, which provides the 
modern answer to Pythagoras's problem of the irrationality 
of such numbers as j/2; and this, or some equivalent theory, 
is needed as part of our analysis of the number system. 
Dedekind defined real numbers in terms of rational numbers 
(i. e. ratios or fractions plq) and there is no difficulty in 
defining the rational numbers in their turn in terms of the 
'natural numbers" 0, 1, 2, — If we can once introduce the 
natural numbers in a satisfactory way, then all is relatively 
plain sailing. 

Detailed examination of the use that is made of the 
natural numbers in mathematics reveals that all that needs 
to be known concerning these numbers is that they constitute 
a progression. In other words, we can generate the entire 
system of natural numbers progressively by starting with 
and repeatedly passing from n, the last number so far 
obtained, to its successor n', i. e. the number n + 1. One way 
of introducing the natural numbers into mathematics — a 
way favoured by both Dedekind and Peano — is to postulate 
the initial number and the successor operation as given, 
and then to adopt axioms which confer on the set of natural 
numbers the structure of a progression. This is precisely 
the function of the well-known Peano axioms for the natural 

(1) is a natural number. 

(2) If n is a natural number, the successor n of R is also 
a natural number. 

(3) No two natural numbers have the same successor. 

(4) is not the successor of any natural number. 

(5) If P is a property such that (a) has the property P, 
and (b) if n has the property P then so also has n, then 
every natural number has the property P. (The axiom 
of mathematical induction.) 

This axiomatic way of introducing the natural numbers 


is completely satisfactory for mathematical purposes. But if 
we do not wish to postulate special concepts such as those 
of zero and the successor operation in addition to the 
general concepts of logic and set theory, we can adopt the 
alternative course of defining a particular progression in set- 
theoretic terms and then using this always to stand for the 
progression of natural numbers — which it can perfectly well 
do since the natural numbers have no mathematically signi- 
ficant features beyond that of being a standard progression. 
In fact we use the standard progression very much as the 
progression 0, 1, 2, . . ., 10, 11, . . , of numbers in the scale of 
ten is used in more intuitive mathematics?. 

The customary definition of the natural numbers as sets 
runs as follows. We denote by the empty set (i. e. the zero 
totality which has nothing at all belonging to it); and if 
a.b,...,k are given entities we denote the set of these entities 
bv {a, b, . . ., fc}. Then we may form the progression of sets 

0, {0}, {0.{0», {0,{0} J {0>{0>}}'--" 

by taking first the empty set 0, and then at each stage 
introducing a further set whose members are all the sets 
previously introduced. The sets of this progression are then 
defined to be the natural numbers, and from this point on 
they are represented by the familiar symbols 0, 1, 2, 3, — 
Such, in brief, is the transformation that mathematics has 
undergone as the result of the criticism to which its founda- 
tions have been subjected in modern times. And such is the 
resulting conception of mathematics that is now presupposed 
as a matter of course in current research in this field. 

References for further reading 

KNEEBONE, G.T., Mathematical Logic and the Foundations of 
Mathematics, London 1963 (especially Chapter 5). 

ROTMAN, B. and KNEEBONE, G. T., The Theory of Sets and Trans- 
finite Numbers, London, 1966. 


Special Relativity 
and some Applications 


The special theory of relativity was proposed by Einstein 
in 1905. During the early days of relativity, his work went 
under a rather different heading from the title we give it 
today. The original paper (Ann. Phys. Lpz. 17) by Einstein 
was called "Electrodynamics of moving bodies". The subject 
has since been called the special theory of relativity, or 
the restricted theory, to differentiate it from his second 
theory, the general theory, which he wrote in 1915. The 
latter theory was a theory of gravitation, based on concepts 
he introduced in his special theory. While the general 
theory could never be regarded as a mere application of 
the special theory, the special theory has, nevertheless, many 
direct applications, so much so in fact, that this theory is 
recognized without question as a corner-stone of modern 

Mathematically, the special theory of relativity is really 
a geometry: a certain kind of geometry of four dimensions 
which connects three spatial coordinates x, y, z and a time 
coordinate t. It is not surprising that it is still important in 
the development of present theories. Relativity is construed 
as an essential part of physics, and all theories must fit in 
with this four-dimensional framework and obey the so called 
relativity principle. 



In order to arrive at this four-diinensional geometry which 
has just been mentioned, and to explain the relativity prin- 
ciple, we must first go back to Newton, or at least his 
conception of space and time. His own framework, on which 
he founded his three laws of motion, supposes the existence 
of an absolute euclidean three-dimensional space and an 
absolute universal time, which is common to all observers 
independent of their position or motion in space. Space and 
time are separate entities. They can be envisaged indepen- 
dently of one another. 

Suppose a particle moves with uniform velocity in one 
dimension (which we may take to be the x-axis). In a 
space-time diagram, the motion is recorded as a straight 
line (Fig. 2.1). 

We can also record a two-dimensional motion in a space- 
time diagram without difficulty. For instance, if a particle 
moves in a circle with constant speed, the space time diagram 
turns out to be a helix (a curve in the shape of a spiral 
advancing along an axis) (Fig. 2.2). 

However, when we try to represent three-dimensional 
motion in a space-time diagram we get into difficulty. We 
should need to go into four dimensions to represent it. But 





Fie. 2.2. 

this is still mathematically feasible; and in fact we often 
do work with a space of four dimensions in this way, 
although it is not possible to represent the situation dia- 
gramatically. The kind of geometry to use between x, y, z, t 
now becomes an important question. 

In classical mechanics, space-time diagrams are merely 
representative diagrams, although we can imagine them 
based on euclidean geometry by a suitable definition of the 
word "graph", without any loss. However, for the purpose 
of bringing the work closer to Einstein, we must be prepared 
to read a little more into the geometrical relationship con- 
necting x and t and to acknowledge that in the end it may 
turn out not to be euclidean. 

But let us first of all see how, on the basis of euclidean 
geometry between x and t in Fig. 2.1, the newtonian picture 
represents, for instance, the motion of light, in one dimension. 
Light travels with a constant velocity of approximately 
3 X 10 10 cm/sec, in a straight line. The magnitude of this 
velocity is commonly denoted by c. It would be useless to 
try to draw a line representing the motion of a light pulse 
on the space-time diagram of Fig. 2.1, since the slope of the 
line would be 

t 1 

x ~ 3 X 10 10 ' 


and this would be so small that the line would be indistin- 
guishable from the axis of x. This, of course, would not 
conflict with Newton's opinion that light travelled instanta- 
neously, but rather than accepting this, since we now know 
otherwise, let us agree to multiply the units in which the 
time t is measured, by c. Then the motion of a light pulse 
moving along the #-axis would be represented by a straight 
line with unit slope [Fig. 2.3 (a)] . We are now measuring time 
T (= ct) in light-seconds (the distance in centimetres tra- 
velled by light in one second). A light-year (the distance 
travelled by light in one year) is approximately equal to 
3- 16 X 10 7 light-seconds. 

I (=cl) 


Fig. 2.3. 

The dotted line in Fig. 2.3 (a) indicates the motion of a 
light pulse emitted in the opposite direction from O. It is 
not difficult to see that a ray of light emitted in both 
directions from O for a finite time interval between T = 
and T = a (say) would be represented as a region of the 
space-time diagram [Fig. 2.3 (b)]. 

If a ray of light shines in both directions for an infinite 
time from T = to T = °o (a = °o) the corresponding region 
is easily obtained [Fig. 2.4 (a)]. This part of the space-time 



diagram is called the future region from O, because all points, 
or rather events (x, T), within it can be arrived at from O 
without having to travel at a speed greater than that of 
light (the slope of all lines from O to points in the region 
is numerically greater than unity). It is, of course, impos- 
sible, from a physical point of view, to have speeds greater 
than that of light. The future region thus constitutes the 
totality of all events which are accessible from O. 

FIG. 2.4. 

In a similar way, the region bounded by light pulses tra- 
velling towards the origin is called the past region from O 
[Fig. 2.4 (b)], and represents the totality of all events which 
may communicate with O. 

The remaining region is called the present region with O 
and is made up of all those events which are neither acces- 
sible from O nor communicable with O. We then have the 
complete picture, built up from light pulses travelling 
towards and away from the origin [Fig. 2.5 (a)]. 

The terminology "past", "present", "future" is used in anti- 
cipation of special relativity, where it is quite common. At 
the moment, the terms past, present, future are relative only. 
They are entirely dependent on the fact that O is at rest 
with respect to absolute space. 

If we consider a three-dimensional space-time diagram, 
representing a two-dimensional motion in the plane of x and 




Fig. 2.5. 

y, the past, present and future regions of this diagram are 
separated by an infinite cone, called the light-cone, whose 
semi-angle has unit slope [Fig. 2.5 (b)]. Any cross-section of 
this diagram through the time-axis brings us back to a two- 
dimensional diagram [Fig. 2.5 (a)]. 

For four-dimensional space-time, the same terminology is 
used, although the light-cone in question is no longer a two- 
dimensional surface but a three-dimensional one. 

Fig. 2.6. 



Before we introduce the idea of relativity, we have a fairly 
good application of the things we have mentioned so far, 
which we might consider for a minute: the past history of 
the universe. Let us choose coordinates by supposing that 
our Milky Way, the nebula which we, in the solar system, 
inhabit, is at the event x = 0, y = 0, z = 0, T = 0. We take 
a cross-section in the ^-direction and represent the situation 
as shown in Fig. 2.6. Consider the light L which travels to 
us from the distant parts of the universe, in fact, from the 
distant nebulae, labelled by the iV's. The distant nebulae 
are understood to be receding from us with speeds which 
vary in proportion to their distance from us.f If we assume 
that the nebulae have always been moving with uniform 
speeds equal to their present speeds, extrapolation of their 
path lines in the space-time diagram will reveal that they 
appear to emanate from one common event in the past. This 
event is at (0, — cE) where H = 4-1 X 10 17 sec (ot 1-3 X 10 10 
years). (It will be recalled that we are measuring the time 
T in light-seconds.) It is as though all the nebulae were 
packed tightly together 1-3 X 10 10 years ago and then moved 
apart as though an explosion had occurred there. It is custo- 

f A galaxy's speed (the word "galaxy" is synonymous with "ne- 
bula") can be estimated by analysing the spectrum of the light 
received from it. The so called "red-shift" — the apparent shift 
of the entire spectrum in relation to known patterns of absorption 
lines (which indicate the presence of certain elements within the 
vicinity of the galaxy) — is interpreted as a Doppler effect, the 
diminution of frequency which is observed when a source is mov- 
ing away from us. The effect is well known in the study of 
sound vibrations. The sudden drop in the pitch of a whistle of 
a train as it rushes past us is a common example of this, involv- 
ing a raised frequency (advance) and a lowered frequency (re- 
cession). Similar principles apply equally well in the case of 
light waves, e. g. an ordinarily green light when travelling with 
a speed of about c/5 would appear blue if travelling towards us, 
and red if travelling away from us. 





mary to say that H, on the basis of this very naive picture, 
is the age of the universe, f 

It is interesting to compare this figure, which is arrived 
at from a purely optical approach, with the age of the earth 
determined from the analysis of radioactive rocks. This 
age is about 2-6 X 10 9 years, possibly higher. (At least on 
the basis of the space-time model the age of the earth is not 
greater than the age of the universe!) 

We return to the earth to describe a certain relativity of 
motion which was known to Newton. So far the measure- 
ment of relative motion has not been mentioned. This is 
precisely where relativity comes in. We shall see that new- 
tonian relativity, as it is sometimes called, fails in its attempt 
to describe relative motion and moving sources of light. The 
reasons for its failure lie in the assumption of an absolute 
space and an absolute time. 

Newtonian relativity is based on the equations 

x = x - Vt, y = y, z' = z, 

which connect two frames of reference: (x, y, z) which is 
fixed in absolute space, and (x\ y', z') which is moving in 
the ^-direction with constant velocity V (Fig. 2.7). 

Newton simply observed that his famous law of dynamics 

force = mass X acceleration 

docs not change its character when we pass from the first 
coordinate system to the second. This is because the law 
is independent of velocity. The essence of this statement 
is that if we perform a dynamical experiment when we are 
travelling with a constant velocity V with respect to the 

t Cosniological theories which envisage a definite point of origin 
usually incorporate II as the age of the universe. There are other 
cosmological theories which envisage no definite origin. A dis- 
cussion of each type of theory is unfortunately beyond the scope 
of this lecture. 

earth, the result we get would be the same as that produced 
by an identical experiment at rest on the ground (assuming 
for the moment that the earth is a good enough representation 
of absolute space: we shall come to this shortly). Velocity, 
dynamically speaking, is not an absolute quantity, but purely 
relative, whereas acceleration is absolute 

Since Newton's second law is referred to as the "law of 
inertia", a coordinate system in which the law is valid is 
called an inertial frame. By the principle of newtonian 
relativity, an absolute coordinate system can be substituted 
by any one of a whole set of inertial frames; the significant 
point is that no one of these frames is more important than 
any other. 

On the basis of newtonian relativity (which it will be 
remembered will fail in its attempt to describe kinematics 
satisfactorily), let us see how the moving frame of Fig. 2.7 

2 Z' 

Fig. 2.7. 

records a light pulse emitted in the usual way from O in the 
fixed frame. Consider the situation in the common x-x' 
direction for simplicity, and set up two space-time diagrams, 
one for the fixed frame and one for the moving frame. 
Fig. 2.8 shows how each frame "sees" the light pulse. 

It is clear that a person moving with the origin O' will see 
the light pulse travelling with a velocity c — V in the forward 
direction and c + V in the reverse direction owing to the 
velocity V of O' with respect to O. 0"s conception of past 





and future will be different from that of O; and, in general, 
any optical experiment whicli is carried out in the moving 
frame will give a different result to that of a similar 
experiment carried out in the fixed frame. Hence the 
principle of newtonian relativity holds for dynamics, but not, 
it seems, for optics. Because of this, and because the motion 
of light is always uninfluenced by the motion of its source.t 

T (=ct) 

T (=ct) 

x = cl 

x'= - <c+V)1 

X'= (C-V)t 

Fig. 2.8. 

it should be possible to measure the earth's velocity with 
respect to absolute space by doing a suitable optical 
experiment on the earth. Such an experiment was carried 
out by Michelson and Morley in 1887. They set themselves 
the task of finding the velocity of the earth with respect to 
the stationary luminiferous aether which in their time was 
supposed to characterize and fill the absolute space (or one 
of the dynamically equivalent inertial frames) devised by 
Newton, which had since become so crucial from the point 
of view of optics. Their experiment, it could be said, was 
designed to detect an "aether wind". 

t This fact emerges in the observation of "double stars", i. e. two 
stars moving in orbits around their common centre of mass. If 
light took longer to reach us from one star than from the other 
(when the first was receding and the second approaching) this 
would give rise to anomalies in the observed motions of these stars 
(viz. under simple velocity addition, circular orbits would appear 
eccentric), and no such anomalies have been found. 

This famous experiment of 1887 was a refinement of a 
previous experiment which Michelson had carried out in 
1880 (at that time the sensitivity of the apparatus he used 
made the result unreliable). Michelson's idea was very 
simple. It was to send two beams of light from the same 
source, fixed to the ground, in two mutually perpendicular 
directions and over equal distances, and then to reflect them 
back again into the same straight line and measure any 
difference in the arrival times of the light waves by observing 
interference fringesf (Fig. 2.9). 

A = glass plate, half-silvered 
on its rear surface. 

M,= mirror, reflecting first beam. 

B = compensating glass plate 
having the same thickness 
as A. This is employed so 
that the two beams have 
equal path lengths in glass. 

M, = mirror, reflecting second beam, 



",-^j„ o 

Fig. 2.9. 

The experiment can be compared with a simple problem 
in relative velocity. It is not difficult to show that it always 

t When two light waves arrive in step with thedr crests together 
they reinforce one another and produce a bright fringe (construc- 
tive interference); when they arrive out of step cancellation takes 
place and a dark fringe is produced (destructive interference). In 
practice any difference in the arrival times of the two beams in 
Michelson's apparatus could be most easily discovered by rotating 
the whole apparatus through 90° when a fringe shift should be 
produced, since the roles of the mirrors in the experiment would 
be changed. 



takes longer to fly an aeroplane at a certain rate up-wind 
and back, over a certain distance, than it does to fly at the 
same rate perpendicularly across-wind and back, over an 
equal distance. (The proof is given in an appendix, but the 
reader should attempt to prove it for himself.) Correspon- 
dingly, for Michelson's experiment, if an aether wind 
prevailed, one light beam must travel up-wind and back 
while the second must travel across-wind and back, and the 
first beam will arrive later than the second. 

The result of the Michelson-Morley experiment was quite 
sensational. The two arrival times were the same. No aether 
wind existed. The experiment was repeated when the earth 
was in a different position in its orbit round the sun and the 
same result was obtained. Again no aether wind. 

The problems posed by the unexpected result of this 
experiment led indirectly to Einstein's theory of relativity. 
Einstein established the idea that the geometry of space-time 
diagrams (so to speak) was non-euclidean. In this way he 
was able to revive the principle of newtonian relativity for 
dynamics and extend it to optics (and electromagnetism) as 
well, and in doing this, the old idea of absolute space and a 
luminiferous aether came to be abolished. 

In short it was supposed that the space-time diagrams of 
Fig. 2.8 were basically incorrect and ought to be replaced by 
identical pictures on the basis of the null result of the 
Michelson-Morley experiment (Fig. 2.10). 

T (=cl) 


Fig. 2.10. 


It was precisely the old transformation rules 
x' = x — Vt, y' = y, z — z 

which embodied the idea of euclidean three-dimensional 
space and absolute time. Instead, new transformations were 
given, in which the time T (= ct') of the moving system was 
no longer the same as T of the fixed system, thus dropping 
the idea that time was something absolute. 

The light pulse x = T had to be the same as the light pulse 
x = T', and x = — T the same as x' = — V . This was 
certainly true. (We follow an argument which Einstein, or 
his contemporaries, might have considered.) 

In addition there was no harm in tentatively assuming that 
the new transformations had an essentially simple character, 
and it was therefore supposed, on the grounds of simplicity, 

x' -T' = h (x - T), (1) 

x + T = I (x + T), (2) 

where k and I were constants to be determined. It was 
reasonable to choose kl = 1 from the point of view of units 
in the two systems. This assumption led to the equation 


.'2 _ T'Z = X 2 - J** 

[multiplying (1) and (2)] and this equation also contained the 
geometrical aspect of the transformation: the quantity 
x 2 — c 2 1- should not change its value on transformation from 
one coordinate system to another. Physically this meant that 
the light pulse must present identical pictures to the two 
frames. (If it had turned out that x 2 + c 2 1 2 were the 
invariant, then the geometry between x and T would have, 
indeed, been strictly euclidean, by Pythagoras.) 





Adding (1) and (2), and subtracting (1) and (2) gives two 
further equations: 

h+ l) X -[ k -l) T > 2T ' = [ k + l 

k ~i u 


fc 2 + l 

x = 


fc 2 -l 

x ~k^l T 

fc 2 + 1 r k 2 - i 
r = — — lr- 



The spatial origin x = must be equivalent to x = Vt (or 
x = VT/c), since it is travelling with speed V in the fixed 
frame (the diagram of Fig. 2.7 still applies). The last 
equation therefore gives 

fc 2 -l 



d h 

k* + 1 



= fi (say). 

A: 2 + 1 c ' 2 k ]/(i- Wc 2 ) 

Thus the rules of transformation are 

x' = (S (x - VT/c), T' = f}(T- Vx/c), (y' = y, z m z), 


x' = p(x-Vt), y' = y, z' = z, f = fi (t - Vx/c*), (3) 


£ = 1/1/(1 - VVc 2 ). 

These are the Lorentz transformations of special relativity. 
It is with these transformations that one associates the 
principle (really a postulate) of special relativity: "no 
experiment can ever detect a uniform motion through space." 
The principle is entirely equivalent to the statement that the 
laws of nature possess the same form in all frames of 
reference in uniform motion relative to each other. In 
particular, the laws of dynamics and the laws of electro- 
magnetism (including optics) must satisfy (and can be shown 
to satisfy) this requirement. 

The equations of transformation (3) themselves produce a 
variety of consequences apart from the properties of 
invariance which we have mentioned. 

(i) When the velocity V between the frames (x, y, z), 
(x' t y', z') is small compared with the velocity of light, as will 
be the case for most terrestrial applications, we find that the 
Lorentz transformations (3) reduce to 

x'=x-Vt, y' = y, z = z, *'==* (0=1). 

These are the old transformation rules, formerly based on 
the notions of absolute space and absolute time (f = t), but 
now to be regarded as approximations. 

(ii) If the equations (3) are solved for x, y, z, t in terms of 
x', y' , z ', f, we find 

x = p(x' + Vt'), y = y', z = z, t = p (f 4- Vx'/c*) 

and this has the same form as (3) except that V is replaced 
by — V. This demonstrates the perfect symmetry between 
the two frames. No one frame is preferred and no one 
frame is fixed in any absolute sense. 

(iii) Simultaneity has no absolute meaning in relativity. 
If two events (*i, y\, z\, t\), (x2, y2, z& t%) are simultaneous to 
an observer, i. e. t\ = t^, they will no longer be simultaneous 
to another observer moving with a velocity V, for it will not 
be the case that t\ = t{, by virtue of the Lorentz transfor- 

t{ = P (h - Vx x lc*), U = fi{h- Vxs/c*). 

These times are unequal. This follows because we have 
xi =|= X2- (If x% = X2 the two events would be the same and 
questions of simultaneity do not arise.) 

(iv) The combination of two successive Lorentz transfor- 
mations is still a Lorentz transformation. Suppose a first 
frame is related to a second by the equations (3): 


x' = 0(x-Vt), y' = y, z' = z, t' = fi (t - Vx/c 2 ), 
iff = 1/1/(1 -V 2 /c 2 ), 

and the second related to a third by a velocity V: 

x" = P'{x'-V't'), y" = y', z" = z', t" = & (*' - V x'lc 2 ), 
F = 1/1/(1 - V'W), 

then, by eliminating x', y', z', t' between the two trans- 
formations, it is not difficult to show that the third frame 
is related to the first frame by the equations 

^ = /5 "( x -P't), y" = y, z" = z, t" = 0" (t - V" xlc 2 ), 

where V" (as it turns out) is not merely the addition of the 
two velocities V+V' as one might at first be tempted to 
imagine, but is given by a formula 

V" = {V + V')l{i + VV'/c 2 ), 

which approximates to V + V' when the velocities are small 
compared with that of light. 

It is interesting to notice that if V = c (or V = c) then V" 
is also equal to c, verifying that the velocity of light can 
never be increased by "adding" a further velocity to it. 

£, e, 


Fig. 2.11. 



(v) Moving measuring rods appear to contract. This is a 
curious phenomenon, but it must be a consequence of rela- 
tivity. Suppose a rod of length I is fixed along the x'-axis 
of the moving S' system (Fig. 2.11) between the points 
x' = Xi' and x' = x-z' (so that x% — x{ = I). 

If the situation is viewed from the frame S at any time t 
say, the length of the rod x-2 — xi as it appears in this frame 
can be found from the following formulae: 

xi = fi (*i' + Vh% t = [it 1 + Vxi'ft?). 

xi = j3xi' + V(t- fiVx'i/c*) = fi (1 - We*) x\ + Vt, 
i. e. 

xi = Xl ' V(l -VW) + Vt. 

x 2 = x 2 ' 1/(1 -V 2 /c 2 ) + VU 

x*-xi = {x 2 ' - xi') 1/(1 - F 2 /c 2 ) = IV{1- V 2 /c 2 ). 

Hence the rod appears to be shortened by a factor 
V (1 — V 2 /c 2 ). This is known as the Fitzgerald contraction, 
(vi) Moving clocks appear to run slow. This is an equally 
curious phenomenon. Suppose a clock is fixed in the frame 
S' (Fig. 2.12) at the point (x', 0, 0). Let t % ' and U be the 
times of two consecutive "ticks" that it makes, and the 
interval between the ticks t% — i\ — p sec. When the 




Fig. 2.12. 





situation is seen from the frame S the interval is no longer 
p sec, for 

h = fi {h' + Vx'/c 2 ), t 2 - P W + Vx'/c 2 ), 
and so 

h-h = P W - W) = P/V (1 - VV<*). (4) 

Hence moving clocks "appear to run slow" by a factor 
1/1/(1 — V 2 lc 2 ): this is called the time dilatation effect. 

The phenomenon has very recently been directly confirmed, 
experimentally, to within a factor of about 2 per cent.f 
The technique adopted did not make use of the ticks of a 
moving clock, but of the very short waves produced by a 
certain moving gamma-ray source (following the discovery 
of R. L. Mbssbauer, which enables us to use certain atomic 
nuclei as very accurate frequency standards). The theoretical 
principles involved are very much the same as if a moving 
clock had been used, but, of course, the success of experi- 
ments on time dilatation depends on the existence of ex- 
tremely precise methods. 

The source was mounted at the circumference of a rotor, 
which was set spinning with a very high angular velocity. 
The radiation from the moving source was received at the 
centre by an absorber. 

Since the frequency of the radiation is inversely propor- 
tional to its period, we have 

where t a , t s are the periods, and v a , v s are the frequencies of 
the received and emitted radiation respectively. The velocity 
V of the source equals cor, where a) is the angular velocity 

f Champeney, D. C, Isaak, G. R., and Khan, A. M., Proc. Phys. 
Soc. 85, 5S3 (March 1965). (Their article also contains references 
to other recent experiments.) 

of the rotor and r is its radius. Thus by the binomial 

v a co 2 r 2 

-==i-a— -. 

V» c z 

(In view of the magnitude of c, higher order terms may be 
neglected.) This theoretical prediction was verified. 

Finally, we consider some dynamical consequences of 
special relativity. 

If a particle of mass m, fixed in a frame S', is moving with 
a velocity V with respect to a frame S, Einstein supposes that 
the mass which an observer in S would measure is not m but 

mtV (i - VVc 2 ) = M (say), 

rather like the time dilatation (4); m, being the mass which 
an observer at rest relative to it in S' measures, is called the 
rest-mass. The quantity M is called the relative mass, and 
correspondingly MV, the momentum which S measures, is 
called the relative momentum. 

In the absence of force and non-dissipative impact, total 
relative mass and total relative momentum are conserved. 
These basic principles of relativity mechanics can be illu- 
strated in the following dynamical problem: suppose two 
spherical perfectly elastic particles move along their common 
line of centres and collide. If M\ and M% are their relative 
masses, and V\ and Vo are their velocities before impact, 
and the same symbols with dashes denote the respective 
quantities after impact, the assumption that total relative 
mass and total relative momentum are conserved yields the 

M ! + M 2 = M x ' + M 2 ', M, Vt + M 2 V 2 = Mi 1 VS + M 2 V 2 . 



If V\ and V% are known, the velocities V\' and V 2 ' after 
impact can be determined, since we have the auxiliary 

Mi = mi/ 1/(1 -F?/c 2 ), 
Mi = mi/1/ (1 - Vf/c 2 ), 

M 2 = m 2 /]/(l -F 2 2 /c 2 ), 
M 2 ' = m 2 7l/(1 -V 2 ' 2 I<?). 

There are two solutions. One of these is obviously V\ = V\, 
Vn = ^Y corresponding to the case of no collision. The other 
is the required solution for collision. 

In the approximation when V\ and V 2 are small compared 
with c, the equation for the conservation of mass gives, by 
the binomial theorem, 

m (1 + &F1W) + m 2 (1 + W 2 ~Ic 2 ) 
= m (1 + hVi'Vc 2 ) + m 2 (1 + W 2 ' 2 lc 2 ). 

If we multiply this equation through by c 2 , we obtain the 
usual equation for the conservation of energy: 

hmi V? + hm 2 Vf = imi Vt* + hm 2 V 2 ' 2 . 

This shows that the conservation of mass and of energy are 
equivalent and related by E = Mc 2 , the famous Einstein 
formula. We can see this also by expanding Mc 2 : 

E = Mc 2 
= mc 2 / ]/(! - V 2 /c 2 ) = mc 2 + hmV 2 + higher terms in Vic. 

rest kinetic 
energy energy 


In special relativity, Newton's second law is interpreted as 

force = - (MV), 



where M is the relative mass. What change does this new law 
make to existing dynamics? Virtually none for terrestrial 
phenomena since the velocities encountered are so small 
compared with the velocity of light. However, the theoretical 
difference the law makes to the motions of the planets round 
the sun, in relation to their calculated positions according 
to newtonian mechanics, is significant in the case of the 
planet Mercury, being within the fringe of what is measurable. 
It can be shown that on the basis of the special relativistic 
form of Newton's second law, the orbit of Mercury is not 
an ellipse with the sun as focus as classical mechanics 
determines, but an ellipse which is slowly but constantly 
rotating about the sun as focus, i. e. an elliptical rosette. 
The rate of rotation, given by these calculations, is found to 
be 1\ sec of arc per century, which is an extremely small 
quantity. Observationally, Mercury does possess this 
annomaly in its orbit, but the rate of rotation is, in fact, 
approximately six times as much as special relativity predicts. 
The added precession is taken to be a consequence and a test 
of Einstein's second theory, his general theory of relativity, 
and this derives for Mercury the observed precession to 
within about 1 per cent. 


To show that it takes longer to fly an aeroplane at a certain 
rate up-wind and back, over a certain distance, than it does 
to fly at the same rate perpendicularly across-wind and back, 
over the same distance. 

Let us suppose that the rate at which the aircraft travels 
in still air is c m. p. h. If the wind-speed is V m. p. h. 
however, it is clear that the aircraft's speed is then c — V 
when it is travelling up-wind, c + V when it is travelling 
down-wind, and V (c 2 — V 2 ) when it is travelling across-wind 
(in either direction). 



If we take the first case when the aircraft flies up-wind 
over a distance of d miles (say) and returns over the same 
distance, the total time of flight is 

d d 


c-V c+V 


In the second case when the aircraft flies across-wind and 
back, the time of flight is 


V(c*-V 2 ) 


the time of the outward and return journies being in this 
case equal. 

Provided F#=0, the first time is always greater than the 
second: the inequality 




c z -V 2 V{c*-V*) 

is valid because c> V (c* — V 2 ). 


Some Properties 
of Integers and Primes 

S. J. Taylor 

Very early in life we learn to count and we build up a 
store of experience of the "whole numbers". We learn that 
we can add them, multiply them and sometimes subtract or 
divide, and that there is always a correct (or unique) answer 
for these operations. Our object in this lecture will be to 
try and formalize some of these notions and to see that many 
of the tricks of manipulation which we learnt to use withoui 
fully understanding them can be justified. 

Laws of algebra 

We assume that we know the set Z = {0, 1, 2, 3, . . . } of 
whole numbers, and that the operations of addition and 
multiplication give a unique answer and satisfy 

a + b = b + a, ab = ba; 

(a + b) + c = a + (b + c), {ab) c = a (be): 
a (b + c) = ab + ac; 

a + = a, a • 1 = a for all a ; 

if a=^0 and ab = ac, then b = c. 

It is worth remarking that we could derive these "laws" for 




Z from a more primitive set of axioms, but this would take 
much too long. Further, one can deduce other simple laws 
from the above. For example, the last law implies that the 
product of two integers a, b in Z can only be if at least 
one of them is zero. 


In learning to count we soon learn that some (finite) sets 
are bigger than others. We formalize this by saying that, 
of two unequal integers in Z one must be larger than the 
other. We write a<Cb for "the integer a is smaller than 
the integer b". The order structure of Z satisfies: 


There is no integer between and 1. 


if a < b, then a + c < b + c 

and ac <. be 
if a < b and b < c, then a < c; 
< a for all a #= 0. 

for all c, 
for all c=t=0; 

In fact many of the results of mathematics depend for 
their validity on the fact that the ordering of Z satisfies a 
stronger condition. 

Axiom. The ordering of Z is such that every non-empty 
subset E of Z contains a smallest element* that is, there is 
an element m e E with m <, x for all x in E (any ordered set 
satisfying this condition is said to be well-ordered)* 

It is interesting that this axiom cannot be deduced from 
the laws of algebra together with the fact that Z is ordered. 
The reason for its importance is that it implies the validity 
of the principle of induction. Before stating this formally 
let us make an easier deduction: 

For suppose that there were at least one c in Z with 
< c <C 1, then the set E of such c would be non-empty. By 
the well-ordering axiom E has a least member m and 
< m <C 1. But then, multiplying this inequality by m, we 
obtain < m 2 < m < 1 so that m 2 also belongs to E and is 
smaller than m. This contradiction establishes the truth of 
the statement. 

Mathematical induction 

We must start by formulating carefully the method of 
induction which we use frequently. An essential ingredient of 
such a proof is that we have a series of statements depending 
on the integer n belonging to Z. If H n is such a statement 

(i) Hj t is true, 

(ii) the truth of H n implies the truth of H rt +i for every 
n in Z such that n > k; 

we want to deduce that H n is true for every n ]> k. For 
example H n could be the statement 


by proving (i) and (ii) for this statement with k = we would 
like to say that it is true for every integer n in Z. If we 
call this method of proof the principle of induction, then 

The principle of induction is valid. 

For. suppose the conditions (i), (ii) are satisfied. Let E 
be the subset of Z consisting of those integers n > k for which 
H n is false. If E is not empty, it has a least element m. 
Then H m is false so, by (i), m=t=fc. Hence m>k, so that 
m — 1 > k, since 1 is the smallest positive integer. It follows 



that m — 1 is not in E so that H m -\ is true and this implies 
the truth of H m by (ii) with n = m — 1 > k. But H m cannot 
be both false and true, so our assumption that the set E is 
not empty is not possible. Hence E is empty, so that H„ is 
true for all n > k. 

We deduced the principle of induction from the axiom of 
well-ordering: it is worth remarking that one could also 
have assumed the induction principle as an axiom and 
deduced that the usual ordering of Z is a well-ordering. 

In order to illustrate the need for care in applying the 
principle of induction, let us "prove" the following. 

False Theorem. All the people in any room have the same 
colour of hair. 

Let E n be the statement that, if a room contains n persons, 
they all have the same colour of hair. Since every room 
must contain a whole number of people it is sufficient to 
prove that H n is true for every n ]> 1 in Z. Applying the 
principle of induction (which we have shown is valid) with 
k = 1, then (i) Hi is true, for in a room with only one 
person he (or she) has the same colour hair as himself. Now 
assume the truth of H n and consider any room containing 
(n + 1) persons. Number these inmates from 1 to (n + 1) 
and first send person number 1 out of the room. The room 
now contains n people and by hypothesis they all have the 
same colour hair. Call number 1 back and send out person 
number (n + 1). The room again has n people in it — all 
with the same colour of hair. But now person number 1 
must have the same colour of hair as persons 2 to n and 
the same is true of person number {n + 1). It follows that 
all (n. + 1) persons have the same colour of hair, and we have 
deduced the truth of J? n + i and established the second 
condition of the method of induction. By induction H„ 
must be true for every n. 



Question. What is wrong with the above proof? 

Hint. Check that condition (ii) is satisfied for every n > 1. 


If a, b are in Z we know that the equation ax = b does 
not always have a solution in Z. When the equation has a 
solution it must clearly be unique: in this case we say that 
b is divisible by a or "a divides b" and write a | b. In general 
it is possible to have a | be without either a\b or ale. 
However, there is a special class of integers a for which this 
deduction is valid and Theorem 2 below will establish it. It 
is more convenient to define them by a different property. 

Definition. A prime is any integer p in Z other than or 
1, such that the only integers a in Z which divide p are 1 
and p. 

Everyone is familiar with the first few primes: 
2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, . . .; 

tables have been prepared listing all the primes up to 
10 7 and some very large numbers are known to be primes. 
However, there is a very neat proof, due to Euclid, of the 

There are infinitely many primes. 

Suppose this is false; then the set of primes is finite and 
they can be listed pi, P2, ■ • •, Pn- Consider the integer 
i = PiP2- • • Pn + 1- This cannot be divided by any of the 

primes pi, pi, . . ., p n - But t is bigger than any of p\, p% p n 

so it cannot be a prime. But any number which is not a prime 
must be divisible by a prime, since the smallest divisor 
greater than 1 cannot have divisors other than itself and 1. 
It follows that t must be divisible by some prime other than 



pu ...,p n and we again have contradicted the assumption 
that these are all the primes. 

We will return later to discuss some of the known (and 
unknown) properties of primes. But first we have to 
continue our more general discussion of divisibility. If a 
and b are in Z {b 4= 0) then we can always divide a by b to 
give a quotient q and remainder r which is smaller than b. 
This can be formulated as a theorem. 

Euclidean Algorithm. Given integers a, b in Z with b>0 
there exist unique integers q, r in Z with 

a = bq + r, < r < b. 

We first prove uniqueness. Suppose q it r i} q 2 , r 2 satisfy 
this relation, then 

so that 

a = bqi + ri = bq 2 + r 2 

& I Qi ~ Q2 I = I n — r 2 I . 

Thus b divides I n — rg I which is less than b, so we must 
have ri — r 2 = 0, whence qi = q 2 . Now let E be the set of 
integers of Z of the form a - bx, with x in Z. E is not 
empty since x = gives the integer a in Z. By the well- 
ordering axiom the set E has a least member r corresponding 
to an integer q in Z. Then a = bq + r and we have only to 
show that 0^r<b. U r ^ b, then a - b (q + 1) would be 
a smaller integer in E contradicting the definition of r. 
Hence the relation is completely established. 

Greatest common divisor 

Given two integers a, b in Z we call cf the g. c. d. of a and 
b if d is a divisor of both a and b, and any x in Z which 
divides both a and 6 also divides d. Thus 

some properties of integers and primes 
d\a, d\b and x I a, re I £> imply .x I d. 


Note that the adjective "greatest" really means that d is a 
multiple of any other divisor. 

Theorem 1. Any two positive integers a, b have a unique 
g. c. d. in Z, denoted (a, b). There are integers u, v (not 
unique, and not both positive) such that 

(a, b) = ua + vb. 

This can be proved by repeated application of the 
euclidean algorithm. (I believe that it is usual nowadays 
to teach people to find (a, b) by factorizing a and b into 
prime factors — this is illogical as the method can only be 
justified if one has already proved this theorem or something 
equivalent.) Applying the algorithm we get 

a = bqi + r u (0<ri<b). 

Now if an integer x divides both a and b it must also divide 
n, and similarly if x divides b and ri it must divide a: it 
follows that (a, b) = (b, n). Repeat the argument on b 
and ri if ri =}= 0, 

b = n q 2 + r 2 , (0 < r 2 < ri). 

As the remainders decrease by at least 1 each time, the 
process must ultimately end with a zero remainder after a 
finite number of steps: 

r»-2 =■ r»-iq n + r n , (0 <, r„<r„_i); 
T n -i = r n q n + h 

It is clear that each of the pairs a, b; b, ri; n, r 2 ; . . ,;r„-i, r n 
have the same set of common divisors: (a, b) = (r n -u r n ) = r n . 





The g. c. d. must therefore be the last non-zero remainder in 
this (finite) process. 

The uniqueness for the g. c. d. follows from the fact that 
if d\, d-z both satisfy the definition for g. c. d., then d\ I lis 
and dz | d\, which implies d\ = cfe. The evaluation of integers 
u, v such that (a, b) = ua + vb can also be carried out from 
the system of equations, by successive substitution giving 
each r,- in turn as a linear sum u<a + v%b: 

r\ = a — bqi = a + (- qi) b, 
rg m b — q 2 n = (— q 2 ) a -f (1 + qiq 2 ) b, etc. 

The existence of a g. c. d. enables us to obtain the most 
important property of primes: 

Theorem 2. For any prime p, if p | ab then p\a or p\b. 

Suppose p | ab but p is not a divisor of b, then the only 
common divisor for p, b is 1, and so (p, b) = 1. This 
means that, for suitable integers u, v 

1 = up 4- vb. 
Multiplying by a gives 

a = upa + oab, 

and it is clear that p divides both terms on the right-hand 
side so that p j a. 

We have already seen (in the proof that there are infinitely 
many primes) that any positive integer which is not a prime 
is divisible by some prime. This allows us to express any 
integer n > 1 in Z as a product of a finite number of primes. 
We can now prove: 

Theorem 3. The expression of an integer n > 1 as a 
product of primes is unique apart from the order of the 

Suppose we have two factorizations into primes: 

li = P\ P2 • • • Pn = q\ 32 . • • q m - 

Then since p\ | qi (q 2 . . . q m ) either p\ | q\ or pi | q 2 (g» . . . q m ). 
by Theorem 2. By repeated application p\ | qi or p\ | q 2 or 
Pi | <Z3 (q4 • • • qm) ; and so on. This means that there is some 
fci for which pi | q,k v But then we must have p\ = q/^and 
this term can be cancelled from the product leaving 

P2 P3 • • • Pn = q2 qs' ... q m ', 

where the right-hand side consists of the remaining factors 
qi. The argument can now be repeated showing that each 
of the factors pi, . . ., p„ occurs among the primes qi. After 
they have all been cancelled there cannot be any q's left. 
Hence n = m, and the factors q,- are just a rearrangement of 
the factors p<. 

The theorems we have proved are all (with the possible 
exception of (a, b) = ua + vb) well known to any schoolboy 
in the sense that he believes them and uses them, even if they 
are not formulated in his mind. Thus all we have done so 
far is to indicate the sort of arguments which can be used 
to justify common arithmetical techniques. It is gratifying 
to know that these are valid techniques since most of us have 
found them very useful in practice. The reader might like 
to evaluate the g. c. d.'s (180, 252), (1001, 7655) by the process 
of Theorem 2 to convince himself that this is a sensible 

Let us now consider some other properties of primes which 
either cannot be proved easily or constitute unsolved 


Distribution of primes 

We have already shown that there are infinitely many 
primes. These can clearly be written clown in increasing 
order of magnitude to give an infinite sequence of prime 

Pi, P2, ■ • •> Pn, — 

If one examines this sequence by looking at a table of primes 
one forms certain impressions: 

(i) the sequence looks irregular if one only considers a 
relatively small part of it at once; 

(ii) prime numbers become rarer among the large integers 
and one feels that if an arbitrary integer greater than, 
say a million, is considered, its chance of being a 
prime is very small; 

(iii) if one counts the number of primes in large blocks, 
say between 2 k and 2 k+1 for k = 5, 6, . . ., one obtains 
a sequence of integers which is fairly regular, and 
does not seem to tend to zero; 

(iv) no matter how far one looks in the table there appear 
to be (occasional) "prime pairs", that is integers n 
such that both n and (n + 2) are primes. 

We can formulate the observations (ii) and (iii) precisely: 
the status of (iv) is quite different — it constitutes a famous 
unsolved problem of number theory and is striking because 
it can be stated so simply: 

Are there infinitely many positive integers n such that 
both n and (n + 2) are primes? 

and yet we arc still nowhere near a solution. Incidentally it 
is possible to prove that any arithmetical progression of the 
form a, a + d, a + 2 d, . . ., a 4- kd, . . ., where (a, d) = 1 (why 



is this condition essential?) contains infinitely many prime 

Since the sequence of primes is completely determined 
(though it is not possible to give a simple arithmetical formula 
which yields only primes, let alone gives all the primes) we 
can define a function by 

/ (n) = number of primes p <L n. 

Since there are infinitely many primes we know that f (n) -> oo 
as n — » oo, but our observation (ii) above leads us to suspect 
that /' (n) is small compared with n for large values of n. 
The density of primes among the first n integers is 

d {n) = 


and can be computed for particular values of n. It can be 
proved without too much labour that 

d (n) — > as n -> oo, 

so that, for very large n, if we pick an integer at random 
among the first n integers it has only a small chance of being 
a prime. Actually, although d (n) -» it does so rather 

slowly, for n = 10 9 , d (n) is approximately ^. 

We observed in (i) that the distribution of individual 
primes among the integers is extremely irregular. However, 
this irregularity "in the small" disappears if we look instead 
at the average distribution given by d (n). The very regular 
manner in which d (n) — >0 as n ~> oo is one of the most 
remarkable discoveries of mathematics. If log n denotes the 
logarithm to base e of the integer n, Gauss already noticed 
that d (n) is very close to 1/log n and that the closeness of the 





approximation appeared to improve as n increased. If we 
measure the goodness of the approximation by 

r(n) = 

1/log n 


r (10 3 ) = 1159, r (10 6 ) = 1084, r (10 9 ) = 1053. 

Gauss conjectured early in the nineteenth century that this 
ratio r (n) -* 1 as n -» °o, but he was unable to prove it. This 
result is known as the prime number theorem. Although the 
result can be stated using only relatively simple concepts, it 
was almost a hundred years before analysis was developed 
sufficiently to provide the tools for a proof. (The first proofs 
were given in 1896 by Hadamard and de la Vallee Poussin.) 
The first proofs used elaborate machinery of complex 
function theory, but more recently so-called "elementary" 
proofs have been devised. These are still too difficult to be 
included in the average honours course for a university 
mathematics degree. 

of "very large". If the Goldbach conjecture is correct then 
it follows easily that every integer n is the sum of at most 
3 primes. 

Nothing in this chapter has been original. I have only 
tried to do two things. Firstly, to indicate how the usual 
familiar arithmetical processes can be justified. Secondly, 
to illustrate that difficult theorems and unsolved problems 
can be stated using very simple concepts. 

References for further reading 

BlRKHOFF, G. and MacLane, S., A Survey of Modern Algebra, 
Macmillan, 1941. 

COURANT, R. and ROBBINS, H., What is Mathematics?, Oxford, 

Goldbach conjecture 

In 1742 in a letter to Euler, Goldbach observed that, for 
every case he tried, each even number, other than 2, could 
be expressed as the sum of two primes. He wanted to know 
if this was true in general. We still do not know today 
though very strenuous efforts have been made on this 
problem. The cause of the difficulty is that primes are 
defined using multiplication, while the problem involves 
addition. However, there has been some progress towards 
attacking the problem in recent years, and it is not hopeless 
to expect a solution in a finite time. For instance Vinogra- 
doff showed that for all very large integers n, it is possible 
to represent n as a sum of at most 4 primes, but his proof 
is an existence proof and cannot yield a precise definition 



P. Chad wick 

1. Introductory remarks 

When interviewing candidates for university admission I 
am often asked the question "What is applied mathematics 
at the university all about?" My stock answer is that 
applied mathematics at the university continues to be largely 
concerned with mechanics, but that we now look beyond 
the motions of particles and rigid bodies to the mechanics 
of materials such as gases, liquids and real solids which 
deform when acted on by forces. This is all very well but 
it gives no idea of what kind of mathematics is involved in 
these more advanced studies, nor does it indicate the enormous 
variety of physical situations which the mathematics is able 
to portray. At the end of a 20 minute interview a short 
answer is perhaps excusable, but today I find myself with 
a generous allocation of time and a large captive sample of 
the kind of people who turn up at interviews. 

In order to try to convey to you some idea of the flavour 
of applied mathematics at the university level I propose to 
discuss in some detail a topic which crops up repeatedly 
during the three-year duration of the typical course. This is 
the theory of wave motion, and the reason for its frequent 
appearance is simply that whenever a material body is 
deformable a disturbance produced locally spreads out 
through the body as a wave. 




Of course waves are a common part of our everyday 
experience. We have all enjoyed generating water waves in 
the bath and idly observing waves by the sea-side. On 
analysing these experiences more closely we realize that 
light waves are involved in conveying the undulations of the 
water surface to our eyes, and sound waves in enabling us to 
hear the noise which accompanies the motion of the sea. 
From time to time we read or hear of earthquakes in which 
waves travelling through the earth's crust cause vibrations 
of the ground capable of demolishing buildings and other 
structures. Recently radio waves from outer space have 
been in the news, while terrestrial radio waves are the means 
of wireless and television broadcasting. 

We are all, then, familiar with waves and in particular the 
crucial role which they play in communications. In your 
case, moreover, this is an informed familiarity because you 
have all had physics lessons in which certain important 
properties of waves have been discussed and explained. 
You know, for instance, that waves can be reflected and 
refracted, can exhibit interference and diffraction, and can 
sometimes stand still as in an organ pipe or on the strings of 
a guitar. But the applied mathematician's approach to waves 
is rather different from that of the physicist in that he is 
interested in building up a general mathematical description 
of wave phenomena rather than in giving so-called physical 
explanations of particular wave properties. The operative 
word here is general. It is characteristic of the mathematician 
that he seeks to place his results in the most general possible 
setting. In this way he is often able to show that seemingly 
diverse ideas have a common basis. Increasing generality 
and the development of unified theories are marked features 
of university mathematics, and I hope that my later remarks 
will show you how these tendencies operate in the particular 
case of wave theory. 

Before turning to some actual mathematics, mention should 
be made of the considerable impact which the study of 



wave phenomena has had on the historical development of 
mathematics. The classical theory of waves is largely the 
creation of the great mathematicians of the eighteenth and 
nineteenth centuries, but during the present century there 
have been revolutionary developments in theoretical physics 
in which the wave concept is of fundamental importance; 
and progress has been made in the study of what are called 
non-linear waves. The best known example of a non-linear 
wave is the shock wave. Shock waves are produced in the 
air by explosions and when an aeroplane accelerates until 
its speed exceeds the speed of sound. These supersonic 
bangs are also much in the news at the present time, but in 
this chapter I shall consider only the more domesticated 
linear waves. 

2. Simple harmonic oscillations 

In order to fix a few ideas I would like first to say 
something about simple harmonic oscillations. Let us con- 
sider three examples. 

(a) Fig. 4.1 (a) shows a particle P of mass m attached 
to one end of a light elastic spring. The other end of the 
spring is fixed and the whole system (i. e. the spring plus 
the particle) is assumed to move in a fixed horizontal straight 
line. Let the spring have natural length / and elastic 
modulus X, and let the displacement of P from its position 
of equilibrium be £. Then the equation of motion of P (i. e. 
Newton's second law of motion applied to P) is 

dH X 

where t denotes time. 

(b) In Fig. 4.1 (b) the particle P (again of mass m) is free 
lo slide on the smooth curve C which is fixed in a vertical 
plane. Let the tangent to C be horizontal at the lowest 
point O and suppose that P is released from rest at a point 


on C close to O. Then if s denotes the arc distance OP and 
R is the radius of curvature of C at O, the equation of 
motion of P is, to a first approximation, 

d s s mg 

"> ~r~ = — ir s> 
dt* R 

g being the acceleration due to gravity. 

\-iumrmrmmrmr- • 9 m 



equilibrium position of P 
Fio. 4.1. Systems executing simple harmonic oscillations. 

(c) The third example (see Fig. 4.1 (c)] concerns a uniform 
circular disc attached at its centre O to a vertical wire the 
upper end of which is fixed. We suppose that the disc is 
mounted in a horizontal plane and is rotated slightly from 
its equilibrium position before being released from rest. It 



then proceeds to execute torsional oscillations governed by 
the equation 

d 2 
/— = -DO, 

being the angular displacement of the disc from its 
equilibrium position. The constants / and D appearing in 
this equation are respectively the moment of inertia of the 
disc about the axis of rotation and the torsional rigidity of 
the wire. 
These three examples have two features in common. 

(i) The motion is completely determined when a single 
variable (f, s or 6) is known as a function of the 
time t. 

(ii) This dependent variable in each case satisfies a 
differential equation of the form 


+ w 2 £ = 0, 


where w is a constant characteristic of the physical 

Thus we can formulate the following statement. 

Whenever the state at time t of a system is specified 
by a single number which is determined by solving 
a differential equation of the form (1), then that 
system is performing simple harmonic oscillations. 

In the above statement the concept of simple harmonic 
oscillations is associated with a purely mathematical idea, 




namely the differential equation (1). This idea is essential 
to all simple harmonic oscillations irrespective of the 
particular physical context in which they occur. Had we 
defined simple harmonic oscillations on the basis of acceler- 
ation being proportional to distance, our definition would 
have had to be worded so as to include linear acceleration 
[example (a)], acceleration along a curve [example (b)] and 
angular acceleration [example (c)], and even then it would 
apply only to the oscillations of a mechanical system. But | 
in equation (1) could represent current in an electrical circuit 
or, say, the brightness of a star at a given time. We have 
here, therefore, an example of the way in which the 
mathematician relates effect occurring in different physical 
situations to a central mathematical idea. In fact we have 
set up what might be called an abstract mathematical theory 
of simple harmonic oscillations which consists of the study 
of the solutions of equation (1). 

As is well known, assuming that o) 4= 0, the most general 
solution of equation (1) is 

f = a cos {oyt + a), 



where a and a are arbitrary constants. In other words, every 
solution of equation (1) can be obtained from the expression 
(2) by assigning suitable values to a and a. For a particular 
oscillation, equation (1) must be supplemented by two 
subsidiary conditions which serve to determine a and a. 
These conditions usually consist in specifying the values 
of f and dtj/dt at a particular time, normally t = 0. In 
physical terms this amounts to stating the manner in which 
the oscillation is excited. 

The basic differential equation, then, together with two 
subsidiary conditions determines a unique solution of the 
form (2). a is called the amplitude of the oscillation and a 



the phase angle. The period, 2ti/co, depends entirely upon 
the nature of the system and not upon the manner in which 
it is made to oscillate. 

3. Oscillations, waves and differential equations 

From what has been said about simple harmonic oscillations 
I hope that you will now understand what is meant by 
saying that, to the applied mathematician, an oscillation 
is something which involves the solution of one or more 
differential equations. The solutions are functions of time 
and they usually have a periodic character like the result (2). 
This leads us to expect that the mathematics of waves, 
which I mentioned earlier, will also be, basically, a matter 
of solving differential equations. Thus the whole subject 
of oscillations and waves derives its mathematical framework 
from the calculus which, after all, has been described (by 
E T. Bell) as the "chief instrument of applied mathematics". 
The mathematical distinction between an oscillation and a 
wave is that whereas oscillations are governed by ordinary 
differential equations, waves are governed by partial diffe- 
rential equations. 

Now before going any further we must be quite sure that 
we understand what these terms mean. In the differential 
calculus we meet functions of a single variable [e.g. f {x)\ 
and study, among other things, properties of their derivatives 
(e. g. dfldx, d 2 f/dx 2 etc.). Later (usually at university) we 
encounter functions which depend upon two or more 
variables, for example q> (x, y). Here we need to know 
values of both x and y before we can determine rp. Now 
this function, <p (x, y), can be differentiated in more than 
one way. We can imagine that y is a constant and then 
differentiate q> as if it were a function of x only — this 
derivative is written dcpldx — or we can treat x as a 
constant and differentiate with respect to y, obtaining in 
this case dcpldy, dcpldx and dcpldy are called the partial 



derivatives of <p f and we can also define partial derivatives 
of the second and higher orders. To take a simple example. 

If cp (.v, y) = x 2 y + xy s , 

dw d<p 

then -J- = 2xy + y* and — = X s + 3 xy*. 

dx Oy 

A differential equation is an equation which contains deri- 
vatives: it is said to be a partial differential equation if it 
contains partial derivatives; otherwise it is called an ordinary 
differential equation. Equation (1) is an ordinary differential 
equation and we shall now expect to set eyes on a partial 
differential equation as soon as we embark on a mathematical 
treatment of waves. 

4. The wave equation 

Having prepared the ground let us now try to arrive at 
this wave equation as rapidly as possible. We shall do this 
by constructing a mathematical representation of a simple 
type of wave. According to the Shorter Oxford English 
Dictionary, a wave consists of "each of those rhythmic 
alternations of disturbance and recovery of configuration in 
successively contiguous portions of a body or medium, by 
which a state of motion travels in some direction without 
corresponding progressive movement of the particles succes- 
sively affected". This is a heroic attempt at the hopeless 
task of embracing in a single statement all the possible 
meanings of the term "wave". However, we single out for 
our present purposes the idea of a wave as a disturbance 
travelling in some fixed direction. 

Consider a "waveform" travelling in a fixed direction with 
constant speed c. We set up two sets of axes with origins O 
and O' as shown in Fig. 4.2 (a) and (b), O being fixed and O' 
moving relative to it along the x-axis with speed c. Thus O' 



moves with the waveform which, relative to the axes with 
origin O', can be represented by an equation of the form 

<P = f(X), 


X being distance measured from O' in the direction of 
propagation. The function / specifies the wave profile. 
rig. 4.2 (a) depicts the situation at time to, and we suppose 
that, at this instant, OO' = a. Then the relation between 

(a) time t, 

ID) lime 

•a + cn-t, 

— X 

Fig. 4.2. Representation of a travelling waveform. 

the distances x and X at time U, is X = x -a. Fig. 4.2 (b) 
shows the position of the waveform at time t {> t ). In the 
interval t — to which elapses between the two states, O' 
moves a further distance c (t — t ) away from O so that, at 
time /, 

X = x~a-c(t- t ). (4> 

Combining equations (3) and (4) we obtain the following 
representation of the travelling waveform at time t: 


<p = f [x — a — c(t — to)}- 



Now the constant a is at our disposal and so we can set 
a = cto: this simply means that O and O' coincide at time 
t = 0. Equation (5) then simplifies to 

<p (*, t)=f(x- ct), 


the notation on the left-hand side indicating that the 
function <p representing the waveform depends upon distance 
x in the direction of propagation and time t. There is no 
need for us here to associate <p with any particular physical 

Equation (6) is the required result. For any function / it 
represents a wave propagating with constant speed c and 
without change of shape in the positive .^-direction. By 
an exactly similar argument a representation of a wave 
propagating in the same manner in the opposite direction is 

<P (x, i)=g(x + ct). 


Now what we are really seeking is a partial differential 
equation which has the expressions (6) and (7) as solutions. 
In order to find this equation we work out some partial 
derivatives of <p. From equation (6), 

<p = f {X), where X = x — ct, and so 



df dX 
~dX dx 

df d<p df dX 
= dX ~dt ~"dX~dt 

d s q> 

d*f d*cp d 2 f 
~ dX*' ~dt* " dX 2 

1 dfy d 2 (p 

c* dt* dx°- 

= — c 





and it can easily be shown that the expression (7) also 
satisfies this equation.f 

We have therefore arrived, by a direct but essentially 
intuitive argument, at what is called the one-dimensional 
wave equation (8). Equation (8) is said to be one-dimension- 
al because <p depends upon time t and a single coordinate x. 

5. Waves on a taut flexible string 

Now that we know what the wave equation looks like we 
can set about showing, in a more or less respectable way, 
how it governs a particular type of wave motion. Probably 
the simplest physical system which admits wave propagation 
is a stretched string, and we proceed on the basis of the 
following assumptions: 

(i) The string is uniform and perfectly flexible; 
(ii) The weight of the string is negligible in comparison 
with the tension; 

(iii) Fluctuations in the tension due to the motion are 

(iv) The string suffers only small deviations from its 
equilibrium position. 

In view of assumption (ii), the equilibrium position of the 
siring is a straight line and we take this to be the x-axis. 
The displacement cp of the string from its equilibrium position 
will, in general, vary with time and from point to point on 
the string. Thus cp is a function of * and t. Consider an 

t It will be observed that the expression (6) satisfies the partial 
differential equation 

_ 1^.— &!£ 
c dt~ dx ' 

which is simpler than equation (8). But this partial differential 
equation is unacceptable because it is not satisfied by the ex- 
pression ("). 



element PQ of the string of length ds and denote by tp and 
tp + dtp the angles which the tangents to the string at P and 
Q make with the x-axis (see Fig. 4.3). Because of assump- 
tions (i), (ii) and (iii), PQ is subject to no forces other than 

Fig. 4.3. Displaced configuration of a taut flexible string. 

the constant tension T acting tangentially at P and Q. The 
equation of motion of the element perpendicular to the x-axis 
is therefore 

gds —^- = T sin (tp + dip) — T sin tp 


= T cos?/; dip, 

where g is the (constant) mass per unit length of the string. 
Dividing by Tds and proceeding to the limit ds — » 0, there 

p d z cp dtp cos tp 

— — — = cos tp — 
T a/ 2 r ds 



R being the radius of curvature of the displaced configuration 
of the string. Finally, we express assumption (iv) in the 

t It should be verified that the terms neglected do not affect 
the limit. 



following more precise form: at all points of the string and 
at all times the angle y is small. Then 

1 _ d*q> 

and with these approximations equation (9) becomes 

1 d s <p 

c 2 dt* 


where c 2 = Tig. Thus the transverse displacement cp satisfies 
the one-dimensional wave equation (8). 

Having shown earlier that the expressions (6) and (7) are 
solutions of equation (8), we can now assert that, subject, 
of course, to the four assumptions which I have listed, 
waves can travel along a taut flexible string in either 
direction without changing their shape and with constant 
speed V (T/q). A system which transmits waves of arbitrary 
form without distorting them is said to be non-dispersioe. 

In order to examine the wave motion of a stretched string 
in more detail, we now take the very important step of 
observing that when we add together the functions / and g 
appearing in equations (6) and (7) we get a further solution 
of equation (8): that is the expression 

q> = f(x-ct)+g(x + ct) 


satisfies equation (8). Now this means that waves can travel 
along our string in both directions at the same time, each 
wave propagating exactly as if the other did not exist. 
Equation (10) tells us that the displacement of the string at a 
given point is obtained by adding together the displacements 
produced by the two waves propagating by themselves. 
Mathematically this result is a consequence of a general 



property of equation (8), namely that if we have a number 
of solutions then their sum is also a solution. This property, 
which is called the principle of superposition, is one of (he 
corner-stones of the mathematical theory of linear waves. 

So far we have not had occasion to specify the form of 
the functions / and g, nor has anything been said about what 
happens when a wave reaches an end of the string. Let us 
now consider a case in which / and g both represent sinu- 
soidal waves by putting 

/ (x - ct) = ha sin{k (x - ct)),g (x + ct) = \ a sin{fc (x + ct)}. 

Here a and k are constants, and the reason for including the 
factor h will appear in a moment. In view of equation (10) 
we can now state that a solution of the wave equation (8) is 

<p = ha [sin{fc (x - ct)}+ sin{fc (x + ct)}\ = a sin foe cos kct. (11) 

FlG. 4.4 Standing wave on a taut flexible string. 

To see what kind of disturbance of the string this represents, 
first take t to be constant; that is take an imaginary snapshot 
of the string at a certain instant. At this instant the string 
is in the form of a sine wave (see equation (11) and Fig. 4.4), 
and we see that there are certain points, called nodes, at 
which the string is permanently undisplaced. Next take x 
to be constant: that is choose a particular point P on the 
string. It follows from equation (11) that P executes simple 
harmonic oscillations about its equilibrium position Po. The 
overall picture which emerges, therefore, is that of a standing 
wave, and we conclude that the superposition of two sinu- 



soidal waves of equal amplitude travelling in opposite direc- 
tions gives a displacement pattern of a stationary (i. e. non- 
progressive) character. 

Referring to equation (11) we see that the positions of the 
nodes are given by sin kx = 0. The non-negative solutions 
of this equation are 

n 2n 

*-*• *• T 

so that the nodes are equally spaced at intervals of nlk. 
Now if we fix the string at a node, then cut it immediately 
to the left of the fixed point and throw away the left-hand 
portion of the string the remainder goes on vibrating exactly 
as before. In other words, conditions at a node are exactly 
the same as conditions at a fixed end of the string. This 
means that a standing wave can be maintained on a string 
of finite length / fixed at its ends provided that / is an 
integral multiple of nlk. But the constant k is at our disposal, 
so it follows (after a little reflection) that a standing wave 
of amplitude a with n + 1 nodes on a string of length I is 
represented by the solution (11) with k = null. Standing 
wave configurations with 2, 3 and 4 nodes are shown in 
Fig. 4.5. 

Although these remarks on standing waves have referred 
specifically to a taut string fixed at its ends, we can state 
the end-result in purely mathematical terms. We have 
shown that solutions of equation (8) satisfying the subsidiary 

g> = when x = and when x = I 

are given by 

nnx nnct 
<p = a sin — 7- cos — ; — , where n = 1 , 2, 3, . . .. 





Now in a university course these solutions would probably be 
found by applying to equation (8) a technique (called the 
method of separation of variables) which yields other 
solutions capable of satisfying different subsidiary conditions. 
In this way an abstract theory for waves satisfying the one- 

First harmonic 
(tunda mental )k =vl\ 

Second harmonic, k = 2*71 

Third harmonic, k = 3w/ 

FlG. 4.5. Standing wave configurations with 2, 3 and 4 nodes. 

dimensional wave equation is built up, and this theory can 
at once be brought to bear on the physical situations from 
which equation (8) emerges. In the remainder of this chapter 
I shall take a brief look at two such situations. 

6. Sound waves 

Sound waves are waves of compression, and since all real 
materials undergo changes of volume when subjected to 
pressure, sound waves can propagate in gases, liquids and 



solids. The situation in respect of solids is rather different, 
and more complicated, than for gases and liquids, however, 
and we shall defer consideration of elastic waves until later. 

Suppose first that we have a tube containing a gas, say air, 
and propagate a wave along this column of air, perhaps by 
striking one end of the tube. Then u, the displacement of a 
typical point of the gas from its equilibrium position, is a 
function of x, distance measured along the tube, and time t. 
It can be shown, by writing down the equation of motion 
of an element of the gas, that If satisfies the one-dimensional 
wave equation 

1 d 2 u _ d 2 u 
c 2 dt 2 ~ dx* ' 

where c, the speed of sound in the gas, is given by 

c* = k/g . 

Here k is a suitable bulk modulus and go is the density of 
the gas in its undisturbed state. By arguments with which 
you are no doubt familiar, the appropriate bulk modulus is 
given by k = ypo, where y is the adiabatic index and po the 
pressure of the gas in its undisturbed state. 

Although waves on a taut flexible string and sound waves 
in a tube are governed by the same partial differential 
equation, they differ in one important respect. The string 
is everywhere displaced at right angles to itself, whereas 
the passage of a sound wave through a gas is accompanied 
by motion of points of the gas in the direction of propagation 
of the wave. Thus we say that sound waves are longitudinal 
waves, while the waves on flexible strings which wc have 
discussed are transverse waves. 

Now, of course, gases are not always confined to tubes, 
and it is a matter of common experience that sound can 



travel through an unconfined body of gas such as the open 
air. The question therefore arises: "What is the equation 
which controls the propagation of sound in the most general 
circumstances?" The answer is the three-dimensional wave 
equal ion 

1 d 2 <p d 2 <p d 2 (f 

~ dx 2 

c* dt 2 

d 2 q> 
dy 2 dz* ' 


w v displacement 

FlG. 4.6 Three-dimensional rectangular cartesian coordinate 


where x, y, z are rectangular cartesian coordinates (sec Fig. 
4.6). The dependent variable <p in equation (12) is a function 
of the four independent variables x, y, z, t, three of which 
are held constant when calculating one of the partial 
derivatives. <p does not represent the displacement of a 
typical point of the gas (which is now a vector quantity), 
but the displacement can be found when <p is known. 

The three-dimensional wave equation (12), besides being 
the basic differential equation in the theory of sound waves, 
enters into the theory of electromagnetic waves, such as 



light and radio waves. There is also a two-dimensional 
wave equation, obtained by deleting the final term of (12), 
which governs the transverse motion of a stretched membrane 
situated, when undisplaced, in the (x, i/)-plane. 

7. Elastic waves 

Finally, we come to elastic waves under which heading, as 
mentioned earlier, we include sound waves in solid materials. 
The property of elasticity has to do with the ability of a 
solid body to deform in response to applied forces and then 
to recover its original form immediately those forces are 
removed. A very wide range of solids exhibit this property 
so long as the applied forces are not too large: examples arc 
the common metals and alloys, rock and concrete. Now 
a solid can be deformed in different ways. Like a gas or 
liquid it can be compressed, and so has a bulk modulus k 
which measures the volume change produced by unit pressure. 
Unlike other forms of matter, however, a solid can sustain 
forces tending to shear, or twist it, and the stress required 
to produce unit shear is measured by the modulus of rigidity 
ft. Thus two material constants, k and //, are needed to 
specify the elastic properties of a solid, and our earlier 
results lead us to expect that these constants, together with 
the density g, will determine the speed with which waves 
propagate through the material. 

The displacement suffered by a typical point of an elastic 
solid during the passage of a wave is a vector quantity and 
so has three components u, v, w (see Fig. 4.6) each of which, 
in general, is a function of the four variables x, y, z, i. It 
follows that wave motion in an elastic solid is governed not 
by a single partial differential equation but by three. These 
equations turn out not to be wave equations, but the follow- 


d 2 u d (du 6d dw\ 

e— = (* + M— I— + — + — \ + f* 

dt 2 
d 2 v 

dx \dx dy dz ) 

d (du dv dw\ ( 

(d 2 u 

d 2 u d 2 u 
dy* dz* 

dt* dy (dx dy dz 

(d 2 v d 2 v d 2 D 
dx* dy* dz* 

d 2 w d , 

e-^ = ( fc + M^- T- + T- 



(du do dw 
dx dy dz 





The solution of equations (13) is clearly a formidable under- 
taking and has been exercising mathematicians for the last 
100 years or so, but we can achieve a surprising degree of 
simplification by supposing that the displacement compo- 
nents depend only upon x and t. Equations (13) then reduce 


d 2 u 


d 2 V 

d 2 o 

9 dt* ** dx* 


d 2 w 
dx 2 


which are three one-dimensional wave equations. 

Now looking for solutions of equations (13) which depend 
only upon x and t is tantamount to considering the possible 
existence of elastic waves travelling parallel to the Ar-axis 
which give the elastic body the same displacement at all 
points on planes perpendicular to the x-axis. Since we 
have come to realize that whatever quantity satisfies the 
wave equation is propagated as a wave, we conclude from 
equations (14) that waves of the assumed type (plane waves) 
can in fact exist. Moreover, we see that there are two kinds 






of elastic waves travelling with different speeds, the two 
speeds being 

a ={(* + ! V)/Q} l/S , c 2 =(///p)V S . 

For all known solids, ci > c-2. 

Since u is the displacement component in the direction of 
the avaxis, the wave travelling with speed Ci is longitudinal. 
Elastic waves of this type are directly analogous to sound 
waves and are called P-waoes (P for primary or for push- 
pull). The waves travelling with speed c& on the other 
hand, are transverse and have no counterpart in gases and 
liquids. They are called S-waoes (S for secondary or for 

We have shown here that an elastic disturbance of a 
particularly simple kind travelling parallel to the jc-axis 
is composed of a P-wave and two 5-waves. In fact this 
result can be generalized and it can be shown that any elastic 
wave travelling in the interior of a solid body is made up of 
a P- and an S-constituent. 

8. Seismology and the earth's interior 

To end on a physical rather than a mathematical note, I 
would like to indicate how the simple facts about elastic 
waves which have just emerged have been used in determin- 
ing in broad outline the internal structure of the earth. 

At the beginning of this chapter I mentioned earthquakes. 
The science which is concerned with the measurement and 
interpretation of earthquakes is called seismology and it is 
the most highly developed branch of physics of the earth or 
geophysics. Earthquakes occur within the outermost 500 
miles of the earth, usually within 50 miles of the surface. A 
particular earthquake causes most disturbance at the place 
on the earth's surface immediately above, but in addition it 
generates elastic waves which travel outwards from the 

source of the eathquake in all directions and can be detected 
at points far removed from the severely disturbed area. 

The elastic disturbance produced by an earthquake 
contains both P- and S-waves which, of course, travel with 
different speeds. Moreover, because the density, bulk mo- 
dulus and modulus of rigidity of the material from which 
the earth is made vary with depth, so do the speeds c\ and 
C2, and this means that the seismic waves do not travel 
through the earth in straight lines but follow curved paths 
as shown in Fig. 4.7. In their journey through the earth 


Earthquake locus 

Fig. 4.7. Paths of seismic waves. 

the waves may bounce off the surface and when this happens 
an incident P-wave may give rise to a reflected P-wave and 
a reflected 5-wave as well. It will be appreciated, therefore, 
that the motion of the ground recorded at great distances 
from an earthquake is very complicated. Nevertheless, this 
is the record of a disturbance which has travelled through 
the deep interior of the earth, bringing to the surface 
information about the nature of the material through which 
it has passed, and it is the deciphering of earthquake 
records (or seismograms as they are called) which has yielded 
much of our present knowledge about the inaccessible interior. 
The task of determining from seismograms the paths along 
which the elastic waves generated by a particular earthquake 
travel through the earth calls for elaborate calculations as 
well as a high degree of skill and experience on the part of 
the seismologist. During the past 70 years, however, a 


network of seismological observatories covering most of the 
continents has been built up and more and more earthquakes 
have been recorded and analysed. From this vast quantity 
ol data, tables have been computed which make it possible 
to reconstruct the wave paths emanating from a particular 
earthquake with great accuracy. The picture which emerges 
from this reconstruction is a striking one. 

Suppose that an earthquake were to occur beneath the 
North Pole. Then over the entire northern hemisphere and 
as far as latitude 15° S. (i. e. to a circle passing through 
central Brazil, Zambia and the north of Australia) the 
arrival of both P and S waves would be recorded. Between 
latitudes 13° S. and 52° S. (a circle through the southernmost 
tip of South America) hardly any ground motion would 
occur, and between 52° S. and the South Pole only P-waves 
would be recorded. The interpretation of this pattern 
resulted from ideas put forward in 1906 by R. D. Oldham 
and later followed up by a number of other seismologists. 
The results obtained by these workers indicate that the earth 
consists of a liquid core surrounded by a thick solid shell. 
The core, being liquid, is unable to transmit S-waves and, 
furthermore, is a much poorer transmitter of P-waves than 
the solid shell. The effect of the core on seismic waves can 
thus be pictured as follows. Imagine the earthquake to be 
replaced by a lamp emitting light of two colours, white 
corresponding to P-waves and red corresponding to 5-waves. 
The core is completely opaque to red light and so will cast 
a shadow at the earth's surface extending from the South 
Pole to some lower latitude. While not opaque to white 
light the core has different transmitting properties from the 
surrounding mantle. This means that it acts as a giant lens, 
focusing the white light towards the South Pole and 
producing a shadow zone at lower latitudes (see Fig. 4.8). 

The presence of a liquid core within the earth had been 
predicted during the nineteenth century from rather different 
considerations. It had been observed, for instance, that on 



descending a mine the temperature increases steadily. This 
rate of increase, if extended to great depths, would imply 
that the central part of the earth must be molten. 


Fig. 4.8. Formation of seismic shadow zones. 

The confirmation of this earlier conjecture represents one 
of the most striking achievements of seismology. The solid 
outer shell of the earth is called the mantle and the depth 
of the core-mantle boundary below the surface (first 
determined in 1915 by B. Gutenberg) is 1801 miles.f 

A further major seismological discovery was made in 1936 
by the Danish woman seismologist Inge Lehmann. I have 
stated that an earthquake occurring beneath the North Pole 
would cast a shadow between latitudes 13° S. and 52° S. In 
fact it is found that the shadow is not quite complete, weak 

t The mean radius of the earth is 3959 miles. 



P-waves being recorded within this zone. Miss Lehmann 
suggested that these waves could be due to the presence of 
an inner core which is a very good transmitter of P-wavcs 
(see Fig. 4.8). This explanation has been confirmed and the 
radius of the inner core fixed at 777 miles. It is widely 
believed that the inner core is solid, but this has not been 
conclusively proved. 

This brief survey shows how the study of earthquakes, in 
conjunction with the theory of elastic waves, yields a broad 
picture of the earth's interior as consisting of three major 
components: an inner core, which is probably solid, sur- 
rounded in turn by a liquid core, and a solid mantle. 

9. Concluding remarks 

In this chapter I have tried to give some inkling of the 
kind of material which appears in university courses in 
applied mathematics and to indicate the habits of thought 
cultivated by the applied mathematician. I hope that 1 
have not utterly depressed you by drawing the curtain aside 
ou some rather fearsome partial differential equations or 
raised your hopes too high by ending with a rather 
spectacular application of wave theory. 

References for further reading 

Bell, E. T., Mathematics, Queen and Servant of Science, G.Bell 
& Sons, 1952 (especially chapter 17). 

Feather, N., Vibrations and Waves, Penguin Books, 1964. 

HODGSON, J. II., Earthquakes and Earth Structure, Prentice-Hall 

Jeans, J. H., Mathematics of Music, in The World of Mathematics 
(ed. J. Newman), vol. 4, pp. 2278-2309, Allen & Unwin, 1960. 

Sutton, O. G., Mathematics in Action, 2»d edition, G.Bell & Sons, 
1957 (especially chapter 4). 


Square Fibonacci Numbers 

J. H. E. Cohn 


It is usually thought that unsolved problems in mathematics, 
and perhaps especially in pure mathematics must necessarily 
be "hard" in the sense that the solution, if one is ever found, 
will involve some essentially new idea or method. (Of 
course, this description of "hardness" has nothing to do 
with length; a "hard" proof may be very short and an 
"elementary" one long and complicated.) In no other field, 
it is alleged, is this more true than in the theory of numbers. 
Shortly after taking my degree, when I said to my tutor 
that I would like to do research in the theory of numbers, 
I was strongly advised against this on the grounds that the 
remaining problems were so hard that it was almost certaiu 
(hat I would fail to solve any, and in the most unlikely 
event of my succeeding I would deserve election to the Royal 
Society rather than a doctorate. 

I feel now that this advice was perhaps rather misguided, 
and I hope to show this by giving an elementary proof of a 
well-known conjecture in the theory of numbers, first proved 
in 1963. As I have said, "elementary" means that all the 
stages of the proof follow by simple reasoning from known 
theorems. At several stages I shall quote theorems or results 
which are well known or easily established. Proofs of these 
are given in the Appendix. 



The problem discussed here is mentioned in Tomorrow's 
Math, by C. S. Ogilvy, on p. 100. (This book by the way 
is a veritable mine of easily understood unsolved problems.) 
Let the numbers u n be defined for all positive integers n, 
by ui = U2 = 1; u„ + 2 = u« + i + u n for n > 1. The first twenty 
or thirty such numbers are easily calculated, after which 
they increase rather rapidly — ueo has thirteen digits. Of 
the first few numbers, it will be observed that u\ = Ui = \ 
and U12 = 144 are perfect squares, and there do not appear 
to be any others, at least for fairly small values of n. It 
had been conjectured that in fact there are no more, and this 
conjecture, although it seemed "reasonable" had so long 
eluded proof that a computer was used to see whether it 
could find any counter-example. The first million numbers 
u n were in turn examined and no further perfect square was 
found. In fact this is hardly surprising; we shall prove in 
Theorem 3 that there are no more at alL 

In the first place we observe that we can extend the 
definition of u n to cover the cases n <, 0, by using the same 
recurrence relation. It is then found that uq = and u_i = 1 
and we shall prove that u n is a perfect square if and only 
if n = - 1, 0, 1, 2 or 12. Secondly, we shall find it 
convenient to consider also the Lucas numbers v n defined 
by vi = 1, vz = 3 and v n *2 — o n + i + v n . These, too, may be 
defined for all integers, positive, negative or zero. 


At this stage it seems appropriate to explain the notation 
which we shall use throughout. All the symbols m, n, x, y, 
etc., which occur, except a and p are supposed to be integers, 
not necessarily positive; the symbol k wherever it occurs 
denotes an even integer, not divisible by 3. The symbol 
(x, y) denotes the largest positive integer dividing both 
x and y exactly. If *=t=0, then we write x\y \i x divides y 
exactly and x <f y otherwise. Finally, if c > 0, we write 



a = b (mod c) if and only if c | (a — b). Thus for example, 
(18, - 27) = 9; (4, - 5) = 1; (0, - 3) = 3; 4 | 12; 12 H 16; 
21 =6 (mod 5); 18 = - 2 (mod 4), and so on. 


We shall need the following result a proof of which 
appears in the Appendix. 

Lemma. 7/ N is positive, and N = 3 (mod 4) then 

^ + 1/2 = (mod N) 

is impossible unless y and N have a common factor. 

From the definitions of u n and v n the following values are 
easily tabulated: 



V n 

— 2 

— 1 


— 1 


— 1 






































Now let a = h (1 + V5) and = i (1 - ]/5) be the roots 
of the quadratic equation 2 = 6 + 1. Then it is a simple 



matter to prove the following sixteen formulae; details are 
given in the Appendix. 

a + = lja0--i, (i) 

u n = 5-'/. fc» - /?»), (2) 

o n = a" + 0», (3) 

2 U„, + n = Un V n + U n D m , (4) 

2 D OT + n = 5 ll m U„ + D m D„, (5) 

u. n = (-l)^u n , (6) 

D_„= (-1)«D„, (7) 

*»-*^ + (-l)*-^, (8) 

D„ + 12 = o« (mod 8), (9) 

2|o m if and only if 3 | m, (10) 

3 + d w if 4|m, (11) 

(un, d„) =2 if 3 1 n, (12) 

(u„, d„) = 1 if 3 + n. (13) 

D A - * 3 (mod 4) if 2 | fc, 3 -f It, (14) 

v m + 2k= - v m (mod d a -) if 2 | fc, 3 + fc, (15) 

U»+2ft = - Um (mod D A ) if 2 | fc, 3 + fc, (16) 

The main theorems 

Theorem 1. d„ = .v 2 is possible only for n = 1 or 3. 

Proof. (1) If n is even, we have by (8) 

D n = D n/j * ± 2 

for otherwise we would have 

(X + D n/j ) (X- Dn h ) = ± 2, 


which is impossible, since x — D n/t and x + D„ !t are either 
both odd or both even; in the former case the product is odd 
and in the latter the product is divisible by 4. 

(2) If n = 1 (mod 4), then either n = 1 or n =|= 1. If n = 1 
we obtain di = 1 = l 2 . If n=f=l» then n — 1 is a non-zero 
integer, divisible by 4. Suppose that 3 r , where r > 0, is the 
highest power of 3 which divides n — 1, Then we may write 
n - 1 = 2 . 3 r . fc, where 2 | fc and 3 + fc. Thus 

D» = D 1+2 3 /- fc 

= (- IKd! (mod D A ) 

= — 1 (mod Da-). 

by repeated application of (15) 

We see therefore that d„4=;v 2 , for otherwise we have 
x z + l 2 = (mod Dk) which is impossible by the lemma, 
since by (14) da- = 3 (mod 4) and, of course, da- and 1 have 
no common factor. 

(3) Finally, if n = 3 (mod 4), then we have exactly as 
above, either n = 3 or n4=3. In the former case D3 = 4 = 2 2 , 
whereas in the latter we may write n — 3 = 2 • 3 r • fc, where 
r ^ 0, 2 I fc and 3 i fc, and so as before 

v n 

= i?3+2.s r .A= (~ l)' r D3= - 4 (mod Da-), 

by repeated application of (15). Hence again d„^=x 2 , for 
otherwise we would have x 2 + 2* = (mod da) which is 
impossible by the lemma, since da- = 3 (mod 4) by (14) and 
so da- and 2 have no common factor. 
This concludes the proof of Theorem 1. 

Theorem 2. d„ = 2 x 2 is possible only for n = 0, 6 or — 6. 

Proof. In the first place, we must have d„ even, and so by 
(10) we can assume that 3 j n. 



(1) If n is odd, then since 3 | n, n must leave remainder 3 
on division by 6, and so remainder 3 or 9 on division by 12, 
i. e. n = 12m + 3 or 12m + 9 for some integer m. Hence by 
repeated application of (9) we have 


D « = D 12m+3 0r °l 

= i>3 or D9 (mod 8) 
= 4 or 76 (mod 8) 
= 4 (mod 8). 

Thus h u n = 2 (mod 4) and so \ o n 4= * 2 for if x is odd then 
a- 2 is also odd, and if x is even x 2 is divisible by 4. 

(2) Suppose now that n = (mod 4). Then either n — 
or n=)=0. If rc = we have d = 2 = 2 . I 2 . If rc 4= then 
as before we may write n = 2 • 3 r • fc, where 2 | fc, 3 H k and r 
is a positive integer. Thus just as before we obtain 

= (-l)s r p =-2 (mod D k ). 

Thus D k divides {v n + 2) = 2 (J d„ + 1); now by (14) v k is 
odd, and so since v n is even, £ v n is an integer, and so Dn 
divides (h d» + 1), that is 

lv n + 12 = (mod d&). 

Now, by (14) Dk = 3 (mod 4) and so by the lemma, wc see 
that h o n 4= x 2 or o„ 4= 2x 2 . 

(3) The remaining case is n = 2 (mod 4), or n = 2 or 6 
(mod 8). Consider first n = 6 (mod 8). Then either n = 6 
or n 4= 6. If n = 6 then v& = 18 = 2 • 3 2 , whereas if n =}= 6, 
we may write n — 6 = 2 . 3 r . fc, where r is a positive integer. 

3 'f fc and now 4| A\ Thus as before o n = ( — l) 3r 06 = — 18 
(mod D k ). Hence D k divides r> n + 18 = 2 (i d„ + 3 2 ), and 


since by (14) Dk is odd, it follows that D/ { divides {h d„ + 3 2 ), 
in other words, 

I v n + 3 2 = (mod D k ). 

Now by (14) Dk — 3 (mod 4), and since 4 | k, it follows from 
(11) that 5 ■{ Dk, in other words 3 and D/< have no factor in 
common. Hence from the lemma it follows that h D n =t= x 2 , 
in other words D n 4= 2.x: 2 . 

Finally, if n = 2 (mod 8) then -n = 6 (mod 8) and by (7) 
o tt = d_„. Thus D n — x 2 if and only if D- n = x 2 , and by what 
has immediately preceedcd this is possible if and only if 
— n = 6, that is n = —6. This concludes the proof. 

Theorem 3. u„ = x 2 is possible only for n = — 1, 0, 1, 
2 or 12. 

Proof. (1) If n= 1 (mod 4), then either n = 1 or n=fcl. If 
n = 1 we have u\ = 1 = l 2 . If rc 4= 1 then as before we may 
write n - 1 = 2 . 3 r . k, where r > 0, 2 | fc and 3-ffr. Thus by- 
repeated application of (16) we have 

U n = "l + 2./.A = (- l) 3 ' Ui = - 1 (mod D k ). 

Thus u.„ + l 2 = (mod Dk), and so by the lemma u n =£ x 2 , 
since as before by (14) Dk — 3 (mod 4) and D/ : and 1 have no 
common factor. 

(2) If n = 3 (mod 4), then - n = 1 (mod 4) and by (6) 
U- n = Un- Thus u„ = x 2 is possible if and only if U- n = x 2 , 
that is if and only if — n — 1 or n = — 1, by the preceeding. 

(3) If n is even, let n = 2 N. Then if u n = x 2 , we have 
by (4) with m = n = N, 

r2 = 

U-2N = UN DN. 

There are now two cases; 




(A) If 3 + JV then by (13) (un, on) = 1 and sot we must 
have un — y 2 and dn = z 2 . Now by Theorem 1 the latter is 
satisfied only for N = 1 and N = 3; the first of these also 
satisfies the former and gives n = 2; the second must be 
rejected since we have taken 3 \ N. 

(B) If 3\N, then by (12) (un, dn) = 2 and sof we must 
have un = 2 i/ 2 and »# = 2 z 2 . Now by Theorem 2 the latter 
is satisfied only by N = 0, 6 or —6. Of these values, the 
first two satisfy the former whereas u_o = — 8 and so 
N = —6 does not satisfy un = 2 y 2 . Hence we obtain exactly 
two more values, namely n = or 12. This concludes the 


To prove the lemma, we shall first need 

Fermat'S Theorem. If p is a prime and p \ z then 

zp- 1 = 1 (mod p). 

For, we observe that if p is a prime then the (p — 1) binomial 
co-efficients pCu p C2, p C$, . . ., pC p _i are all divisible by p, 
since if 1 <L r <, (p — 1), pC t is an integer which equals 

P(p-l)(p-2)...(p + l-r) 

and since p is prime, none of the factors in the denominator 
divides p since r <. (p — 1). Thus 

f Proof in Appendix. 



(x + 1)p = xp + pCi xp~ 1 + . . . + PC p -i x + 1 

= xP + 1 (mod p) 

{x + 1)P - (x + 1) = xP - x (mod p), 

and by repeated application of this we obtain 

z p — z = (z - \)p - (z - 1) = ... = IP — 1 =0 (mod p). 

Thus p divides zp — z = z (zp -1 — 1) for every integer z, and 
so if in addition p \ z wc have that p divides zp -1 — 1, or 
zP- 1 = 1 (mod p) as required. 

Proof of the lemma. We wish to prove 

// N is positive and N = 3 (mod 4) tf/ierc ac - + ,i/ 2 = 
(mod N) is impossible unless y and N have a common 

Proof. Suppose if possible that iV = 3 (mod 4), that N is 
positive, that (y, N) = 1 and that N divides x* + y*. If N 
is a prime, write N = p; if N is not a prime we may write N 
as a product of primes p\, P2, • • ., Pk some of which may be 
repeated. Now since N is of the form 4 m -r 3, it follows 
that not all these primes p r can be of the form 4 m + 1 (since 
the product of two numbers of this form is also of this form), 
and clearly since N is odd all these primes p r are odd. 
Thus one at least must be of the form 4 m + 3. Write p for 
this prime (or for the smallest such if there be more than 
one). Thus in any case, we have since p divides N, p = 3 
(mod 4), (p, y) = 1 and x 2 + y 2 = (mod p). 

Now since (p, y) = 1 it follows from Euclid's algorithm! 
that there exist integers a and b such that ay — bp = 1- 
Let z = ax. Then 

f See, for example, Chapter 3, p. 36. 



z- = a= x* 

= — a 2 y 2 (mod p) 

m - {bp + l) 2 (mod p) 

= — 1 (mod p), 

and so in particular p •f z. Also p = 3 (mod 4) and so 
h (p — 1) is an odd integer. Thus 

zP -i = ( Z 2)V* (p-D = - i ( m od p) 

which is impossible by Fermat's theorem. 
This concludes the proof of the lemma. 

Proof of (1), (2) and (3). We know that a = h (1 + V*>) 
and = J (1 — J/5) and so 

a + = 1; a£= - 1 (1) 

follows by simple calculation. Also we have by calculation. 

ui = 5-'/* ( a - 0) = 1 

u 2 = 5-'/t (a 2 - £ 2 ) = 1. 

Now suppose that for 

re = m, and re = rei + 1, u„ = 5 -1 /* (a n — n ). 

"m+2 = "m + 1 + Urn 

= 5-Vt{a m (a + 1) -/?"' (/?+ 1)} 
= 5- , Ma m+2 - m+s ), 

since a and /? are the roots of 0- = + 1. The proof of (2) 
for re > 1 now follows by induction. To prove (2) for re <L 0, 
we observe that 


uo = 5-* (a - 0°) = 0, 

u_i = 5- , /t( a - 1 -y?- 1 ) = 1. 

Now suppose that for re = — rre and re = — m — 1, 
u» = J-"A (a n - 0"). 


u_ OT = u-m-i + M-m-2. and so 

U-m-2 = M-w — re_ W) _i 

= 5' 1 '* {a'" 1 - a-" 1-1 - 0~ m 4- y?-" 1-1 } 
= 5-*/* {a-" 1 " 2 (a 2 - a) - /5~ m - 2 (0* ~ 0)) 
= 5 -1 /* (a _m - 2 - /?~"»-2) 

since a and satisfy 2 — = 1. Again the result follows 
by induction. This proves (2); (3) follows in just the same 

Proof of (4) ared (5). We now have 

Um o n + u„ v m = 5- l '*{(a m - m ) (a n + n ) + (a n - n ) (a* + m )} 
= 2 . 5- 1 /' (a m+n - m+n ) = 2 a m + n 

5 u„, u„ + 9* o n = {a m - m ) (a n - n ) + {a m + m ) (a n + /?") 
= 2 (a m+n + m+n ) =2o„ 1 + n . 

Proo/ of (6) and (7). 

u_„ = 5- ,/ *(a- n -/?-") 

= 5- , / ! {(-^)n-(-a)»} by (1) 

- (- I)"" 1 v n , 

D_ n = a _n + 0~ n 

= (-0)n+ (-„)» 

= (-D n P„. 


Proof of (8). Putting n = m in (5) we have 

2 v 2m = 5 u m ! T D m 2 

and putting n = — m in (5) we have, in view of (6) and (7), 

4 = 2D =(-D'*- 1 (5u /n 2 -D w s). (i) 

Thus 2 D2m + ( — l) m 4 = 2 p TO 2 , which on rearrangement be- 
comes (8). 

Proof of (9) and (10). Putting m = 12 in (5) we obtain 




2 D n + 12 = 5 Uj£ U n + D12 P„ 

= 720 u„ + 322 p„ 
On + 12 = 360 U* + 161 p„ = p„ (mod 8). 

Thus since 03, dq, P9 and P12 are the only even p r for 
1 ^ r ^ 12, it follows that 2 | p OT if and only if 3 | p w . 

Proo/ of (11), (12) and (13). Suppose that 4 j m, that is 
m = 2 M, where M is even. Then by (8) 

v m = dm 2 - 2, 

and so 3 | o m would imply that 3 | (p ro + 3) = pa< 8 + 1, and 
this is impossible by the lemma, since clearly (1, 3) = 1. 
This proves (11). 

To prove (12) and (13) we observe that if d = (u„, p„) then 
d 2 divides 5 n* 2 - o n 2 = ± 4 by (i). Hence d = 1 or 2. Also 
by (i) u n and v n are either both odd or both even, and so 
d = 1 if o n is odd, and d = 2 if o n is even, that is by (10). 
d = 2 if 3 I n and d = 1 otherwise. This concludes the proof 
of (12) and (13). 

Proof of (14). If 2\k and 3 + fc, then fc = 2, 4, 8 or 10 
(mod 12), and so by (9), 

Dk — 1>2, D 4t *>8 or D10 (mod 8) 
= 3, 7, 47 or 123 (mod 8) 
= 3or7 (mod 8) 
= 3 (mod 4). 

Proof of (15) and (16). If k is even, then by (4) putting 
n = m = k, 

U2k = u k v k = (mod p a ), 

and by (8) v»k = o/. 2 — 2 = - 2 (mod P*), 

and so using (4) and (5) we obtain 

2 Dm* 2 fc = 5 U m Iff* + D m D2k = ~ 2 D m (mod D/t), 


2 U OT + 2/j = Um»2A + U2AOm= ~ 2U„, (modD/t), 

and we have (15) and (16) on dividing by the factor 2, which 
is legitimate since by (10) d* is odd. 

Proof of statements on p. 76. Suppose that ab = x 2 and 
(a, b) = 1. Then suppose if possible that there exists a prime 
p which occurs in the factorization of either a or b (say a) 
raised to an odd power, 2 k — 1 say, where k i> 1. Since p | a 
and (a, b) = 1 it follows that p t b, and so p 2 k ~ 1 1 ab but 
/>-' fc -I ab. Hence p 2 k_1 1 a: 2 but p 2 k % x 2 , which is impossible 
since if p tk ~ l I x 2 then p k | x and so p 2fc |^; 2 . Hence every 
prime factor o:' either a or b occurs in the factorization of a 
or b respectively raised to an even power and so we must 
have a = A* and b = B 2 . 

Now suppose that ab = x 2 and (a, b) = 2. Then 2 | a, 2 | b, 
so 4 I x 2 hence 2 I x. Let a = 2 a, b = 2b' and * = 2 *'. 



Then a' b' = x' 2 and now (a', b') = 1. Thus by the previous 
result, a = A s and b' = B 2 and so a = 2 A 2 and b = 2 B 2 , 
which was to be shown. 


Digital Computers 
and their Applications 

M. Levison 

In an age of major technological advances, there can be few 
with such far-reaching potential effects as the development 
of the electronic digital computer. In a span of less than 20 
years, the computer has emerged from the laboratory to 
become an extremely powerful tool touching the everday 
livcs of a large part of the population. The pay-slip of the 
factory-worker, the rate-bill of the householder, the house- 
wife's grocery list, the schoolboy's examination marks — these 
are but a few of the many items which at some stage in their 
existence may be processed by computer. 

One of the less happy results of the computer's phenomenal 
rate of progress has been the air of magic and mystery 
which has grown up around it. Yet there is no fundamental 
reason why this should be so, for the basic principles 
underlying it may be readily understood by the layman 
with little knowledge of mathematics and none at all of 

The present article is in two parts. The first of these 
describes what a computer is and explains the purpose of 
its various individual sections; the second outlines briefly 
a few of its many current applications, to give the reader 
some impression of the extent of the computer's vast potential. 






Let us begin with a description of the computer and the 
five basic constituent units of which every computer is built. 
In order to describe the part which each of these units plays 
in a calculation, it will be convenient to draw an analogy 
with a clerk working in a company pay office, whose weekly 
task is to calculate the amount of pay due to each of the 
company's employees, given certain items of data, such as 
the number of hours each employee has worked that week, 
his rate of pay per hour, his income-tax code number, and 
so on. 

Consider then the fundamental activities which the clerk 
must undertake in carrying out these calculations. The first 
operation in evaluating each individual's pay will probably 
be to multiply his hourly rate of pay by the number of hours 
worked in order to arrive at his gross pay for the current 
week. Thus the clerk will have to be able to carry out simple 
arithmetic operations, such as addition, subtraction and 
multiplication. For this purpose he may be provided with 
a pencil and paper, or preferably, for greater speed and 
accuracy, with a desk calculating machine. 

Another task which the clerk must perform is to control 
the "flow" of the calculations, that is to say, he must decide 
which arithmetic operations to carry out at each stage, on 
which data the operations are to be performed, and what to 
do with each of the results he obtains. This he does by 
acting in accordance with a set of instructions which he has 
previously been given, perhaps by bis boss on his first day 
in the pay office, or perhaps (in effect) by a school 
mathematics teacher some year earlier. This set of instruc- 
tions the clerk has throughout the calculation to retain in his 
"memory", either literally or artificially, by committing them 
to a piece of paper which he can read whenever necessary. 
In the course of the calculation, certain intermediate results 
may be produced which will be required for further compu- 
tation at a later stage. These, too, must be recorded in the 
(natural or artificial) memory of the clerk. A further task 

which the clerk may be called upon to perform is that of 
taking certain very elementary decisions. In order to achie\p 
greater flexibility in devising the plan of a calculatioiiv it 
may be found necessary that the list of instructions given to 
the clerk should contain certain branch points. He may, for 
example, be told that if an employee's tax code number 
exceeds 50 then he is to obey one set of instructions, while 
if it is less than or equal to 50 then he is to obey an alternative 
set. The clerk must therefore have the ability to decide the 
relative size of different numbers. 

The items described above are in fact the basic ingredients 
of any calculation, and each of the clerk's activities has its 
counterpart on the computer. Like the pay clerk, the com- 
puter, too, must be able to carry out arithmetic operations. 
For this purpose all computers are provided with what is 
called an arithmetic unit. Pairs of numbers can be sent to 
this unit, and their product or their sum or their difference 
or their quotient can be obtained from it. It would be quite 
possible to use the arithmetic unit of a computer in exactly 
the same way as one uses a desk calculator, with the human 
being sitting at the machine sending numbers to the 
arithmetic unit and receiving the appropriate results back 
again. It would, however, be very uneconomical to build 
an expensive unit capable of adding and subtracting numbers 
in times of the order of 1 millionth of a second (on the most 
recent machines), if the operator is only going to be able to 
insert the numbers at a rate of about one every 5 or 10 
seconds. The computer is therefore provided with its own 
control unit and memory unit. Into the memory unit the 
list of instructions (called a programme) is placed by the 
human operator before the calculation begins. The control 
unit then takes these instructions and carries them out one 
at a time, sending numbers when appropriate to the 
arithmetic unit and dispatching the results received to their 
correct destinations. The memory unit may be used not 
only for storing the instructions which the computer is to 



obey, but also for storing the intermediate results which the 
human being might write down on paper. Like the human 
pay clerk, the computer has an element of discriminatory 
ability. It, too, can compare two numbers stored in its 
memory, decide which is larger (or whether they are equal), 
and follow alternative paths of instructions according to the 
result obtained. 

So far then we have encountered three of the basic, 
constituent units of the computer. There are two others to 
consider. The first of these is its input unit, which enables 
the computer to receive information, both instructions and 
data, from the outside world. Information to be communi- 
cated to a human being must be passed to him by way of 
his sensory organs, most frequently his eyes or ears. In the 
case of the pay clerk, he will receive his instructions and 
data either in the form of the spoken word, or in the shape 
of a document which he may read, or possibly in braille 
which he can detect by touch. The sensory organs of a 
computer usually take the form of devices capable of 
detecting photoelectrically the presence or absence of holes 
punched in a card or a length of paper tape. The hole 
patterns, which may be interpreted as binary digits, are 
transmitted by the device to the computer memory. 

Now the human being is taught at an early age to recognize 
certain groups or patterns of sounds which make up his 
native language and to assign to them certain specific 
meanings. By using this language, any other person may 
readily communicate with him. In the case of the computer 
certain patterns of digits have a built-in meaning as instruc- 
tions or data determined by the computer's electronic circuits. 
These form the basic "language" of the computer and we 
may communicate information to the computer by expressing 
it in this language. 

The fifth of the basic units of the computer is its output 
unit, whereby it may convey information, for example, the 
results of its calculation or details as to how the computation 



is progressing, back to the outside world. For this purpose 
the computer may perhaps be made to punch further holes 
in cards or in paper tape, or to operate some form of electric 
typewriter. A functional diagram of a computer illustrating 
the relationship between the basic units is shown in Fig. 6.1. 




Renin d 

CUMl'tML i 

l«HUl <l WttMI.C 


'Uaill •• 

Maim *.mfr*'t 

I'lG. 6.1. Diagram showing the basic units of a computer, 
and their interconnections. In addition to the connections 
shown, all units receive control signals from the control unit. 

To gain some idea as to what is involved in expressing 
the plan of a calculation in the language of the computer, 
we return once again to our analogy with the pay clerk. 
Once he has calculated the gross pay of any employee, his 
next task is to perform a more elaborate series of calculations 
resulting in the deduction of PAYE income tax. Now the 
actual means whereby such deductions are assessed are very 
complicated, and to understand them fully involves a careful 
study of successive Finance Acts. In order, therefore, that 
the system may be operated by employers not expert in tax 
affairs, the Inland Revenue prepares a special booklet 
comprising a set of instructions and also some tables of 
figures. For each employee, there is provided a tax code 
number, determined by the allowances to which he is entitled, 
and a card divided into eight numbered columns. 

The instructions, expressed in a slightly simplified form, 
begin as follows: 


(1) Write the employee's gross pay for the current week 
in column 2 of the card. 

(2) Add this to the entry (made the previous week) in 
column 3, writing the sum back in column 3. 

(3) Obtain from the tables the amount af tax-free pay to 
date corresponding to the particular date and tax 
code number, writing this in column 4. 

(4) ... 

The instructions continue in a similar manner until ultimately 
at the end of the calculation, the amount in column 7 is the 
tax due for the current week, while a further subtraction 
gives the employee's pay after tax has been deducted. 

What has in fact been achieved is that the elaborate 
process of evaluating the income tax due has been broken 
down into a series of simple operations, such as recording 
a number on a card, adding two numbers together, and so 
on. The pay clerk has in effect been programmed by the 
Inland Revenue in much the same way that a computer must 
be programmed by a human user. The computer is capable 
of certain basic elementary operations, such as recording 
a number in some given cell of its memory, adding together 
two numbers stored in its memory, and so on. More 
complicated operations must be broken into steps such as 
these. Nevertheless, using only these very simple operations, 
one can build up the most complicated and complex 
calculations. Indeed, it is possible to programme the 
computer to manipulate not only numbers, but also symbols 
and alphabetical data, for, as we shall see later, it is quite 
simple to represent the latter by some numerical form of 

The reader may perhaps be tempted to ask whether the 
human user might not be able to perform the calculations 
himself in the time taken to write so detailed a programme 



of instructions. The analogy with the pay clerk again 
provides the answer. It is clear that he need only be given 
instructions on pay calculations for one of the employees 
Thereafter, he may evaluate the pay of any other employee 
simply by repeating the same set of calculations, but with 
fresh data applicable to that employee: a different rate of 
pay, another tax code number, and so on. In fact a great 
many calculations contain repetitive parts like this. Fur- 
thermore, it is found that there are many sequences of 
computer instructions, carrying out some particular operation 
which occurs over and over again. One may, for example, 
want to calculate the square root of a number many times 
during a calculation, and in many different calculations. 
Exactly the same sequence of computer instructions can be 
used each time. Sequences of calculations of this type are 
known as subroutines. Over the course of time a great set 
of these subroutines may be gathered together, thus sub- 
stantially reducing the time taken to programme a new 

By this stage the reader should have at least a rudimentary 
picture of the structure and the functioning of a digital 
computer. Let us now turn to considering some of the 
purposes for which computers are being used. 

The bread and butter applications of computers are to be 
found mainly in the commercial world. We have already 
seen that a computer may be used to carry out pay-roll 
calculations, and there are many other programmes that fall 
under the heading of business applications. Computers, for 
example, are now used by many large business concerns to 
keep an inventory of their trading stock. This inventory is 
updated each day or each week, and the computer reports 
when the quantity of any stock item falls below a critical 
level so that the goods may be re-ordered. In some cases 
the comptitcr is also provided with such details as the cost 
of each item, the time delay in receiving ordered goods, and 
the average frequency with which the item is requested by 



customers. It is then programmed to determine the best 
balance between a heavy cash outlay in re-ordering stock 
and loss of trade through not having the stock available 
when called for. 

Banks, too, have begun to use computers to file and update 
details of their clients' accounts. Evidence of this is to be 
found in the appearance during recent years of curiously- 
shaped figures at the bottom of many cheques. These figures 
are carefully designed to allow them to be read directly 
into a computer by a special piece of apparatus, without the 
need for preparing punched cards or paper tape. At the 
present time the majority of banks use these figures to 
record only the code number of the branch. At least one 
major bank, however, has reached the stage of recording not 
only the branch number but also the cheque number and 
the customer's account number, while the amount of the 
cheque is also inserted once it has been used and returned 
to the bank. The entire processing of the cheque may then 
be handled by computer. It is only by adopting techniques 
such as this that banks will in the future be able to keep 
pace with the growing volume of business which they will 
be called upon to undertake. 

Calculations of the type just mentioned are often referred 
to as data processing. Data processing operations are 
characterized by the fact that a large quantity of data is 
fed into the computer, a large quantity of results are output, 
but only a fairly simple series of calculations are performed 
on them. Thus, although any computer whatever would be 
capable of performing the calculations, these would be done 
most efficiently on a machine with extensive input and 
output equipment and only a simple arithmetic unit. In 
contrast, there are many other problems in which the amount 
of input and output is light while the quantity of computation 
is very large. For such problems it may be more efficient 
to use a computer with a more complex arithmetic unit but 
with fewer input and output devices. Larger establish- 



ments with varied problems may, of course, possess 
computers having the best of both worlds. Such computers 
will carry out efficiently all types of problems. 

Computers are also used extensively for calculations in 
the scientific and engineering fields, including both calcu- 
lations hitherto undertaken by human endeavour, and 
calculations which on account of their size could not 
previously have been contemplated. 

To give an example, in recent years a number of individuals 
have been launched both by the USSR and the USA to orbit 
the earth in artificial satellites. None of these launchings 
could have taken place without the assistance of the 
computer at a great many points. A particularly important 
task for the computer occurs in the moments immediately 
after the rocket has been launched, for it is then essential 
to determine as rapidly as possible, whether the satellite 
will enter a satisfactory orbit. For this purpose, the position 
of the rocket must be carefully observed at short intervals 
of time during the first few seconds of flight. These 
positions may then be substituted into the appropriate 
equations, and details of the orbit calculated. Were this 
task to be undertaken by a team of mathematicians, a great 
many hours or even days would elapse before the solution 
was known, by which time the unfortunate astronaut might 
have landed far from the nearest rescue ship in the middle 
of the Pacific Ocean. With the help of a computer, however, 
details of the orbit may be determined within seconds, and 
the astronaut ejected from his rocket if the orbit is not 

A situation of a similar type occurs in weather prediction, 
where barometric observations, etc., may be used to set up 
equations describing the motion of the atmosphere. These 
equations must, however, be solved within an hour or two if 
the results are to have any value in forecasting weather 
conditions before they actually occur; so that, here again, 
the use of a computer is essential. 



In addition to their use in numerical calculations, com- 
puters have also been applied to the translation of languages. 
The reader may well wonder how a machine designed 
primarily to perform calculations on numbers may carry 
out operations on the words of a language. This may be 
quite simply achieved, however, by coding the linguistic data 
into a numerical form. We may, for example, choose to code 
letter A as 01, B as 02, C as 03, . . ., Z as 26. Then a word 
such as CAT would be represented by the number 030120, 
and BEAST by 0205011920. Once coded in this way, the 
words may obviously be stored in the computer memory 
and may be compared for equality or alphabetical order, 
just as numbers are compared for size (indeed, if we adopt 
the coding system just proposed and if the numbers are 
aligned by the left, then the alphabetical order of the words 
is the same as the numerical order of the numbers represen- 
ting them). 

A rudimentary form of translation might proceed as 
follows. In the computer memory there would be stored a 
long list or "dictionary" of, say, French words. In adjacent 
memory cells would be stored for each of these its English 
equivalent. Given a text in French, each word would be 
read into the computer, looked up in this French dictionary 
(possibly by comparing the text word with every dictionary 
word in turn until an equal one is found, though there are 
much quicker methods), and the English equivalent output. 
In this way, the computer would produce a word for word 
translation of the given text. 

Now as a translation this effort would be of little use. 
It is apparent, however, that the machine might go much 
further; for one can readily store in the computer memory 
not only words but other coded information, such as 
grammatical details, and so on. Opposite each French 
word in our dictionary, we might have not only its English 
equivalent, but also some information telling us whether this 
particular word is a noun or an article or an adjective or 



a verb. Instead of reading the text a word at a time and 
printing its translation, the computer might read an entire 
sentence at a time and look up each of the words in the 
dictionary. It might then make various adjustments to these 
words according to the grammatical information. If in 
French, for example, the computer detected a noun followed 
by an adjective, it could be made to invert the equivalents 
to their more normal English order. 

Close study of the translation problem reveals a number 
of other difficulties many of which have yet to be overcome. 
They form, however, an interesting field of current research. 

Apart from the actual translation of languages, computers 
have also been used in the furtherance of literary studies; to 
help in the preparation of the detailed word-indexes which 
literary scholars call concordances, to collect from texts 
statistical data which may be used in the study of authorship, 
and to assist in the jig-saw-like problem of reconstructing 
small fragments into manuscripts. 

Another problem area to which computers have been 
applied is the playing of games. The games concerned fall 
into one of two groups. For some, such as "noughts and 
crosses" and "Nim", a complete solution is known, either in 
the shape of a set of rules ensuring a win for one side, or 
(which amounts to much the same thing) in the form of an 
exhaustive analysis of all possible positions. Computer 
programmes to play this type of game present few difficul- 
ties, and are of little real interest. 

Rather more interesting are those games, such as draughts 
and chess, for which no winning system is known and for 
which the vast number of possible positions makes exhaustive 
analysis wholly impractical. The draughts-playing pro- 
gramme of A. L. Samuel is a good example. This plays by 
looking at all possible positions two or three moves ahead, 
and pursuing even further those positions in which develop- 
ments (i. e. exchanges or jump moves) are imminent. The 
resultant positions are then evaluated in terms of features 



such as material advantage, mobility of pieces, and so on, 
the computer selecting the move which leads to the best 
score available, subject to the assumption that its opponent 
will attempt the opposite. 

Two other noteworthy aspects of this programme are that 
it retains in its memory details of board positions encountered 
in play for use in later games, and that it can, in the light of 
success or failure, modify the weights attached to the 
different features used in evaluating positions. Thus, after 
a number of games, the programme "learns" to play a better 
game. Indeed such was the improvement shown in the skill 
of Dr. Samuel's programme, that it was able to defeat one 
of the leading players in the United States. 

The behaviour of this programme bears a close resemblance 
at a number of points to that of a human being. It is in 
fact one of a group of programmes which have been written 
with the intention of imitating human behaviour. Apart 
from their considerable entertainment value, the more serious 
motive behind them is an attempt to discover more about 
the working of the human mind. To mention but a few 
examples, programmes have also been written to play chess, 
to prove theorems in logic and pure geometry, and so on. 

Naturally in the limited space available this article has 
been able to discuss only a mere handful of the very many 
fields to which digital computers have been and are being 
applied. It is hoped, however, that this will have been 
sufficient to give the reader some idea of the vastness of 
their scope and potential, so that he may be better able 
to appreciate the impact which computers will make on the 
world of the future. 

Reference for further reading 

Samuel, A. L., Some Studies in Machine Learning using the Game 
of Checkers, in Computers and Thought (ed. E. A. Feigenbaum 
and J. Feldman), McGraw-Hill, New York, 1963. 


The Isoperimetric Problem 

H. G. Eggleston 

The isoperimetric problem can be described as follows. 
We consider a closed curve which does not overlap itself 
and which lies in a plane. Fig. 7.1 (a) shows examples 
of the types of curve that we have in mind; Fig. 7.1 (b) 



Fig. 7.1. 




shows examples of curves that we do not consider. The 
curve will divide the plane into two parts of which one is 
enclosed by the curve and the other is not. The points in 
the part of the plane enclosed by the curve form a set which 
we refer to as the "enclosed set". We can now state the 
isoperimetric problem which is, If we know the length of 
the curve, what is the largest possible value for the area of 
the enclosed set? In other words we consider the class of 
all such closed curves which have a given length and we 
wish to describe, in some form or another, that curve which 
encloses the largest possible area. In very simple cases we 
can actually calculate the area; for a square, equilateral 
triangle or circle of perimeter length L the enclosed set has 
area L 2 /16, L 2 /12]/3, Dl±n respectively. But to solve the 
isoperimetric problem we must consider curves for which it 
is quite impossible to calculate the area of the enclosed set, 
and we need quite different techniques. 

This problem is an interesting one for several reasons. It 
illustrates in dramatic form the difference between asking 
a question and finding the answer, or rather of establishing 
the validity of the answer. The problem had occurred to 
the Greek geometers of over 2000 years ago but its complete 
solution was not discovered until the 1880's. This solution, 
which is that the curve must be a circle, was known to the 
Greeks, but it is one thing to know what the solution of a 
problem is and quite another to establish this solution in an 
adequately convincing form or, as we say, to prove it 
mathematically. Another reason for being interested in 
this problem is that although formulated in mathematical 
terms that are relatively concrete and close to physical 
reality, yet its solution depends upon abstract concepts that 
were not defined until comparatively modern times. Again 
the techniques developed to solve this problem are applicable 
in a wider context for attacking a whole range of problems, 
called by the generic title of Extremal Problems. Finally, 
the problem is one of a very general type in mathematical 



work, that of establishing some relation or connection 
between a property of a set (in this case the enclosed set), 
and a property of its frontier, (in this case the curve). 

One of the difficulties, which the Greeks came across, with 
this problem, was that of providing a suitable mathematical 
formulation. One of the difficulties in mathematics, as in 
most other intellectual disciplines, is knowing where to start. 
The Greeks started from geometrical concepts. Their basic 
elements were geometrical ones; points, lines, planes, and so 
on. Greek geometry, enunciated with such remarkable 
lucidity by Euclid and his fellow workers, dominated 
mathematical thinking and teaching in post-Renaissance 
Europe even until the beginning of the present century. The 
greatest mathematician of all time, Isaac Newton, felt that 
he must write mathematics in a geometrical form. He made 
discoveries by using his own invention, the calculus, and 
was able to prove results that would have been quite beyond 
the capacity of Euclid and the Greek geometers, but having 
once got his results, he felt that he had to clothe them in 
geometrical language. Somehow this made the most 
revolutionary discoveries respectable. At a later stage 
when Gauss had discovered non-euclidean geometries, he 
dared not publish these terrible heresies, because of a well- 
founded fear of the ridicule of his fellow mathematicians. 

Now, however, all is changed. Euclid is no longer a 
textbook at school or at the university, and we do not publish 
mathematical papers in geometrical form. Geometry of the 
euclidean type is definitely out, and non-euclidean geometry 
is equally definitely on the way out. We now base the 
whole of pure mathematics on the concept of sets. Sets of 
what? Well, not sets of anything very specific. Mainly 
sets of numbers and sets of sets of numbers and sets of sets 
of sets of numbers, and so on. Indeed, the idea of a numbeT 
can itself be defined in terms of sets. We need to make 
various precise assumptions about the nature of a set. When 
these have been made we believe that the whole of pure 



mathematics can be deduced from them by entirely logical 

It may, of course, be arguable at any one time as to what 
is logical and what is not, but there is generally a consensus 
of opinion which will agree as to what is valid and as to 
what is not valid, even though these ideas of validity change 
from time to time and what we consider to be a valid proof 
now, may well be regarded by our mathematical successors 
as being merely a primitive approximation to a logical 

Our problem, then, has to be assessed in terms of mathe- 
matical concepts. We have used the expressions closed 
curve, enclosed set, area, length and also the idea of largest. 
Each of these concepts is to be translated into mathematical 
language in a rigorous way. When this translation has been 
done we can only operate mathematically on our new concepts. 
It is no longer permissible to use any of the physical pro- 
perties from which we abstracted to secure the mathematical 
analysis. Once we have defined our mathematical ideas 
we leave the physical world behind completely; we live 
exclusively in a mathematical world which has nothing in 
common with reality and in which it is completely inad- 
missible to do anything except mathematics. It may be 
wondered that in this situation, mathematics ever produces 
results which have any contact with the physical world at 
all. If the arguments that we use are divorced from 
physical reality, it does seem rather surprising that the 
results should yet have physical applications, but we can 
look at the matter rather in the light of a translation from 
one language into another. If we regard physical pheno- 
mena as being some sort of a language and a mathematical 
abstraction of these phenomena as being some sort of a 
translation of them, then it is feasible that the process of 
translation into mathematics and of the use of logical 
arguments in that mathematics should produce results which 



can be re-translatcd into an interpretation of the physical 
environment from which they were first abstracted. 

One of the geometers who worked on the isoperimetric 
problem in the last century was Jacob Stciner. He produced 
a number of solutions whose validity was later challenged 
by Weierstrass on the ground that they contained an 
unproved assumption. Although there was this gap in his 
arguments, Steiner's ideas have proved to be very important 
in this and allied problems. 

Basically Steiner's idea was as follows. Firstly, be believed 
(correctly) that the solution to the problem was a circle. 
Suppose that we denote the circle whose perimeter length 
is L by C then the area enclosed by C is L 2 /4jt. Consider 
some other closed curve of length L, say K, enclosing a set of 
area A. If we can show (L 2 /47i) ^> A whatever K is considered, 
then we shall have solved the isoperimetric problem. But 
this is difficult; we cannot establish this inequality directly 
because C and K may be of very diverse shapes and 
quite different from one another. Instead of making the 
comparison from K to C directly, let us try to go by small 
steps from K to C. Suppose that we could find a new 
curve K* of length L enclosing an area A* so that A* ^ A, 
and suppose further that if K is not a circle then we can 
find K* so that A* > A (a strict inequality), then shall we 
not have solved the isoperimetric problem? Steiner thought 
that we should, but actually there is a gap in his argument. 

If we were able to establish the situation described above, 
then we should have proved the assertion "If K is not a 
circle then K is not a solution to the isoperimetric problem"'. 
This is logically equivalent to "If the isoperimetric problem 
has a solution then that solution is a circle" but it is not 
logically equivalent to "The circle is the solution of the 
isoperimetric problem". One has to assume that the iso- 
perimetric problem actually has a solution; that is to say 
one has to assume that there actually is a curve that has 
length L and, of all such curves, encloses the largest possible 



area. This may not be the case at all. In fact there are 
similar questions which do not have a solution in this 
sense. For example, the Kakeya problem asks "What is 
the least area of a plane set, which has the property that 
a segment of unit length can be rotated completely round 
inside the set through 360° back to its original position?" 
In fact given any positive number s one can find a plane 
set of area less than e having this property, but one cannot 
find such a set with s replaced by zero. Thus the above 
question is wrongly phrased. In the Kakeya problem there 
is no set of least area. 

Perron has illustrated the gap in Steiner's argument in 
a striking form. We know that there is no largest positive 
integer or whole number. If we assume for the moment 
that there is such a largest positive integer then we can 
prove that it is 1 by an argument similiar to Steiner's. We 
consider any integer n and transform n to n 2 . Now n 2 ^ n 
and if n=j=l then n 2 > n. Hence no integer, except possibly 
1, is the largest. Therefore 1 is the largest possible integer. 
This nonsense follows from the false assumption that there 
is a largest possible integer. Steiner's assumption that there 
was a solution to the isoperimetric problem was not false but 
did need to be established as correct. In fact this has been 
done and we use the concept of compactness to determine 
a wide class of problems which we know have solutions. 

Although Steiner's argument had this gap, his idea of 
approach has proved a very fruitful one in many fields. We 
shall now assume that the isoperimetric problem has a 
solution and consider how Steiner put his programme into 
effect. For simplicity we shall restrict our consideration 
to convex curves. They will be closed curves in which the 
direction of the line of motion (or tangent) rotates in one 
and the same sense as we describe the curve once. Figure 
7.2 (a) gives examples of convex curves, Fig. 7.2 (b) exam- 
ples of non-convex curves, and Fig. 7.2 (c) a curve which, 
although it satisfies this condition on the rotation of the 



(ai Convex closed curves 

(hi Non-convex closed carves 

(ci A curve that satisfies Ihe 
condition on the sense ot 
rotation ot Ihe directed tangent 
Out is not closed 

Fig. 7.2. 

tangent, yet does not qualify for consideration, since it is not 
closed. A property of a convex curve that we shall assume 
is that a straight line either (i) does not meet it at all, or (ii) 
meets it in one point, or (iii) meets it in two points or (iv) 
meets it in a segment (Fig. 7.3). 

Given the convex curve K of length L Steiner's aim was 
to find a second curve K* of length L enclosing an area that 



Fig. 7.3. 

was strictly larger than that enclosed by K, unless K was 
actually a circle. How was he to define K* from K? He 
considered the properties of circles, particularly those which 
characterized circles amongst all convex curves, and then 
used the fact that K, if non-circular, did not satisfy these 
properties. One such property that circles possess is that 
J he angle in a semicircle is a right angle. We can phrase 



this as follows: if C is a circle and a, b are two points on 
C such that the two arcsf into which a, b divide C are of 
equal length then any point c on C is such that Z. abc = \n. 
Moreover this property characterizes circles, if a, b are given 
fixed points and c a variable point such that ^ abc = I n, 
then c lies on a circle of which a and b are diametrically 
opposite points. Suppose then that K is not a circle and 
that p and q are two points on K dividing it into two arcs 
of equal length then there must be a point r on one of these 
arcs such that ^L prq =j= \ n. (See Fig. 7.4.) 

Denote the two arcs of K with end points p and q by 
A'i and K z . Suppose that K x is that arc which contains r, 
and that it is itself divided into arcs K u , £12, by r, where 
£]] has end points p, r and K i2 has end points r, q. If we 
move q along the line pq to q x and r on the circle centre p 
radius pr to r x so that distance n q, = rq, then we can 
construct arcs congruent to K n joining p, rj and congruent 
to Ki 2 joining n, q h These two arcs form an arc Kf with 
a total length equal to that of £, and enclose/with segment 
pqu an area (say T) which differs from the area enclosed 
by segment pq and K x by exactly the amount that the area 
of the triangle pn q x differs from that of prq. If ^ prq > h 71 
take qi nearer to p than q; if ^ prq <h n take g, further away 
from p than q. In either case if q x is near q the area of the 
triangle pn q\ exceeds that of the triangle prq. 

Thus the arc Kf has the sort of properties that we require, 
but of course we still have to modify the arc K z . We can 
do this in an exactly similar way provided that we can find 
a point s on it with ^ psq =|= J n and such that if ^ prq > \n 
then ^_ psq > In whilst if ^ prq < %n then ^ psq < hx. This 
could certainly not be done if K 2 was a semicircle. So 
Steiner put a preliminary argument in to cope with this 

t By arc we mean a connected part of a closed curve. It need not 
be an arc of a circle. Any two points on a closed convex curve 
divide the curve into two arcs. 



possibility. He observed that he could choose the points 
p, q on K so that (i) they divided K into two equal length 
arcs as before and (ii) neither of these arcs was a semicircle. 
To see this we have to show that if K is a convex curve 
with the property that when we divide K into two equal 
length arcs by two points p, q, then one at least is a semicircle, 
then K itself is a circle. Take any two points p, q dividing 
K into two arcs K\, K 2 of equal length. Suppose K 2 is a 
semicircle. If K\ is a semicircle then K is a circle. If K\ 
is not a semicircle there is a point t on K\ that does not lie 
on the circle of which £2 is a semicircle (Fig. 7.5). Let u 

be the point of K such that tu divides K into two equal 
length arcs. Denote these arcs by £3, K,\. By hypothesis at 
least one of K3, K 4 is a semicircle. Each of these arcs has 
an arc in common with £2 and therefore, if it is a semicircle, 
lies on the same circle as K 2 does. Thus t lies on this circle. 
This is a contradiction, which shows that our original 
assumption was false, and K was in fact a circle. 

Thus Steiner's subsidiary argument is proved. Given that 
K is not a circle we can find two points p, q dividing K into 
two equal length arcs K\, K 2 neither of which is a semicircle. 


Next, in order to make sure that we can pick r on K t and 
s on K 2 with Z.prq, </psq both greater than In or both 
less than hn, we select that one of K u K 2 that with segment 
pq encloses the larger area (if both enclose the same area take 
either) and reflect it in line pq. Suppose that we had selected 
Kt and that £1 and its reflection K\' in pq form the closed 
curve £3 (see Fig. 7.6). Take r on K t as before, r' its reflection 

Fie. 7.6. 

in line pq on K% f . Move q along pq and define £,'* from K\' 
in precisely the same way that K x * was defined from £,. 
The two arcs K x * and K\* together form a closed curve K 4 
that encloses a larger area than that enclosed by K and is 
of length L. 

K 4 need not be convex. It may fail to be convex at only 
the four points p, q, n, n'. We can replace part of the 
curve K 4 near each of these points by a straight segment 
(the operation at p is shown in Fig. 7.7) to produce a convex 
curve £5 whose length is less than or equal to that of K A 
(because a straight line segment is the shortest distance 
between two points) and enclosing an area at least as large 
as that enclosed by K 4 . 

If the length of K 5 is L u expand K 5 by a similitude with 
the ratio LlL\ [transforming the point (x, y) of the plane 
into the point {{LlU) x, {LIU) y)] to obtain a new curve K*. 





Fic 7.7. 

Finally, K* is of length L, is convex and encloses an area 
larger than that enclosed by AT. Thus Steiner's procedure 
is completed. 

An alternative method devised by Steiner was to symmet- 
rize the set concerned. We make use of the fact that any con- 


Fig. 7.8. 

vex curve encloses a convex set, and that such a set meets a 
straight line in either a point, a segment or in no points at all. 

Consider a convex curve K and a straight line X (see Fig. 
7.8). For each point p on X denote by m p the line through 
p and perpendicular to X. This line meets the set enclosed 
by K in (i) no points, or (ii) one point, or (iii) a segment. 
Take on m„ in case (i) no points, in case (ii) the point p 
and in case (iii) the segment whose midpoint is p and whose 
length equals the length of the segment in which m p meets 
the set enclosed by K. 

We do this for every p of X. The totality of all the 
constructed points forms a set that is enclosed by a convex 
curve denoted by K . The proof of this is omitted. It can 
also be proved that the length of K ? does not exceed that 
of K, and, indeed, that the length of K x is strictly less than 
that of K unless K x is itself symmetric about some line parallel 
to X (in which case K x is a translation of K). Here "symme- 
tric about a line" means "coincides with its reflection in the 


I shall not prove any of these statements. But if we assume 
that they are true, and if we assume that there is a solution 
to the isoperimetric problem, then the properties imply the 
following. If, say, the curve T is a solution of the isoperi- 
metric problem and we take a particular direction then 
there is a line l lying in the direction 6 such that f 
coincides with its reflection in l . Take two perpendicular 
directions and let the corresponding lines /,,, l v+x/t intersect 
at O (see Fig. 7.9). Then all the lines l pass through O. 
For, if this were false, select one l not passing through O 
and let the line through O perpendicular to this l Q meet f 
in p and q. Since F is symmetric about Iq one of p, q is 
nearer to O than the other, say p nearer than q. Since r 
is symmetric about l v and l v+3t/i , is convex and contains q, it 
also contains the rectangle with centre O and one vertex at 
q. But p is an interior point of this square which is 
impossible as p lies on f itself. This is a contradiction. 



Fig. ?.9. 

Hence Iq must pass through O and Iq is simply the line 
through O in the direction 0. T is symmetric about every line 
through O. 

But then T must be a circle, centre O. Otherwise there 
are two points h, j on T whose distances from O are not 
equal (see Fig. 7.10). Take the line I which is the internal 
bisector of angle jOh. r is symmetric about /. Thus t passes 
through points j, h', h, j' in some order, where h' and j' are 
the reflections of h, j in I. This contradicts the convexity 

Fig. 7.10. 



of r, but I shall not give a detailed proof of this. Thus 
again we have a solution of the isoperimetric problem. 

These solutions of the isoperimetric problem by Steiner 
are similar in the sense that they depend upon a modification 
of an extremal curve. The procedure is: 

(i) assume there exists a solution, 

(ii) modify the solution curve in some such way that, 
using the fact that it is a solution, we can identify 
it geometrically. 

Many geometrical problems can be tackled in this way. 
The power of the method lies in the vast range of possible 
modifications available, the limitation of the method lies in 
the fact that there are only a few curves that can be identi- 
fied geometrically. The method is also going to be difficult 
to apply if there is more than one solution, i. e. if in addition 
to a circle there had been some other curve that solved the 
isoperimetric problem, our modification would have had 
to be chosen so as to distinguish these two curves from all 
other convex curves. This is usually not possible. The 
modifications used by Steiner were obtained from two 
properties of circles: 

(i) an angle in a semicircle is \ n, and, 

(ii) a circle coincides with its reflection in any line 
through its centre. 

One of the advantages of Steiner's approach to the 
isoperimetric problem is that it gives us a method by which 
many other problems can be tackled. An entirely different 
approach is due to the Danish mathematician Tom Bonnesen. 
The idea behind this approach is to consider not one convex 





curve so much as a family of associated convex curves. The 
existence of this family enables us to construct a more 
elaborate mathematical structure, which makes more of 
mathematics available and leads to a simpler solution. 

Consider a convex curve C of length L (see Fig. 7.11). 
About each point p of C as centre construct a circle of 

Fig. 7.11. 

radius x. We get a family of circles whose envelope we 
denote by C x . As x varies, taking positive values, we get 
a family of curves and it is this family of curves that we 
operate with instead of C itself. To begin with, each curve 
C x is convex. I shall not prove this. Next if C is of 
length L and encloses an area A then C x is of length 
L + 2 tix and encloses an area A + Lx + nx 2 . These results 
are because C is convex. They may be proved by prov- 
ing them first in the particular case when C is a polygon, 
and then approximating to a general convex curve C 
by convex polygons. The proof when C is a polygon is 
easy because then C x is bounded by segments equal and 
parallel to the sides of C and by some circular arcs. For 
example in Fig. 7.12 the heavily shaded area can be put 
together to form a circle of radius x. Their total area is 

tix 2 . The lightly shaded areas have an area Lx and the 
dotted portion is the set enclosed by C, hence of area A. 

I shall not attempt to validate the approximation argu- 
ment. The area of C x is A + Lx + tix 2 . This expression 
is quadratic in x. We consider it for all values of x both 
positive and negative. This is a typical mathematical 
manoeuvre. The negative values of x do not correspond 
lo any very obvious geometrical entity, but we can use them 
to throw light on the behaviour of the quadratic function for 
positive values of x. Consider the equation 

Fig. 7.12. 
nx 2 + Lx + A = 0. 

When does this have real roots? Completing the square, the 
equation is 

L\* L 2 A 

x + 


4ti 2 n 


Thus we have real roots if and only if — > A. But this 






last inequality is the assertion that the area of C does not 
exceed the area of a circle of perimeter L, i. e. it is the 
assertion that the solution of the isopcrinietric problem is 
a circle. Thus we have only to show that for any convex 
curve C nx 2 + Lx + A = has real roots in order to solve 
the isoperimetric problem. When x is large nx 2 + Lx + A 
is large and positive. Thus the equation nx* + Lx + A = 
will have real roots if and only if we can find some value 
of x (which would be positive or negative) for which 
nx 2 + Lx + A is negative or zero. Now there exists in the 
area bounded by C a circle of largest possible radius. Let 
this largest possible radius be r. Then we shall show that 
nx 2 + Lx + A is negative or zero when x = — r. We call 
r the inradius of C. 

We again consider only the case when C is a polygon. 
The general case can be deduced by an approximation 
argument that we shall not consider. The largest circle 

modification ol C 

Fig. 7.13. 


inside the convex polygon C either touches two opposite 
parallel sides, or it touches three sides which, when extended, 
form a triangle containing C. We deal with this last case. 
The other case is slightly easier and we shall not consider it. 

The process is illustrated in Fig. 7.13. 

Let I be this incircle of C and let O be its centre. The 
three points of contact of I with C divide C into three arcs 
de, ef, fd as shown. Suppose that the sides of C containing 
d, e, f respectively when produced form the triangle ghj, 
then we modify C as follows. Each side of C that lies in 
arc ef we move parallel to the line gO towards O until it 
touches I or lies on a line touching J. Similarly each side 
of C that lies in arc ed we move parallel to jO towards O, 
until it lies on a line touching / and finally each side of C 
in arc df we move parallel to the line hO towards O until 
it lies on a line touching /. 

The final figure is not convex but it contains 7 and thus has 
an area exceeding nr 2 . Suppose that the sides of C in order 
are of length L\, Lo, . . ., L/ k . and that L,- is distant p,- from O 
where i = 1, 2, . . ., k. Then if, as before, C encloses an area 
A and is of length L, 

k k 

A = ShLi p t , L = 2 L it 
i l 

where the first equation is obtained by dividing C into 
triangles each with O as a vertex and two adjacent vertices 
of C as the other two vertices. When we modified C the 
side of length Lf was moved a distance (p; — r) until finally 
it was distant r from O. In this motion it described a 
parallelogram whose thickness perpendicular to the side of 
length Li was (p, — r). Thus this parallelogram was of area 
Li (pi — r). The total loss of area in the modification was 


2 Li (pi — r) and the final area exceeded nr 2 . Hence 
A - 2 Li (p { - r) ^ nr 2 , 



i. e. 

i. e. 


Ttr 2 - rl 

tL > nr 2 , 
- A<0. 

Thus «** -f Lx + A is negative or zero if x = — r. Hence 
it has real roots and thus finally the circle is a solution of 
ihe isoperimetric problem. 

1 his method is fundamentally different from Steincr's. 
We here make no assumption that there actually is a solution. 
Our method of proof shows that there is a solution at the 
same time that it identifies the circle as being a solution. So 
far by this method we have not shown that the circle is the 
only possible solution. But we can also do this. 

Firstly there is a smallest circle 7i containing C (again 
we assume C is a convex polygon and omit the approximation 
argument needed in the general case). Either there are two 
diametrically opposite points of h that belong to C or there 
are three points of 7i that belong to C and form the vertices 
of a triangle which contains Ou the centre of l\ t in its 
interior. We consider this second case only; the first case 
is similar but rather easier. Suppose the three points of 7i 
on C are d, e, f. Select points j, g, h lying in the interior of 
the angles less than n, ^L dOi c,ZeOif,Z fOi d. Move each 
segment of C in arc de parallel to 0\ j away from 0\ until 
ii lies on a line touching fr. Perform similarly on the seg- 
ments of C in arcs ef, fd using g, h instead of j. Let R be 
the radius of If, R is called the circumradius of C. The 
final figure includes Ii and encloses an area not less than 
ttR 2 . Hence 

A + 2Li(R- Pi ) >nR\ 

1. C. 

7i R 2 - RL + A < 0. 

Thus n X s + Lx + A is non-positive when x = — R. Hence 
the distance apart of the roots of n x 2 + Lx + A = is not 
less than R — r. Hence 


(V A\\ 
2 >R-r. 


Squaring gives L 2 — 4ti A i> tz 2 (R — r) 2 . 

Thus L 2 = 4 7i A only if R = r, i. e. C is a circle. 

We can prove the isoperimetric inequality for the class 
of all convex polygons by an entirely different argument. 
We can proceed by an argument which although it modifies 
the length and the area concerned, does so in such a way 
that the quantity L 2 — 4-tiA (which is known as the 
isoperimetric deficit, L being the length and A the area of 
the convex set), is reduced. The procedure also reduces 
the number of sides of the convex polygon. If follows that 
if we can show that the isoperimetric deficit is positive for 
a triangle, then it will follow that it is positive for all 
convex polygons. This argument is an attractive one because 
it makes use of the polygonal nature of the sets concerned. 
It will not be possible to apply it to convex sets which are 
not polygons but its application to polygons is particularly- 

Consider then a convex m-sided polygon, say P, with 
length of perimeter L and area A, we shall assume of course 
that m is greater than 3. (See Fig. 7.14.) We shall construct 
a fc-sided polygon, where k is less than m, say Pi, of length 
of perimeter L\, and area A\ such that 

L 2 - 4nA>L l 2 - ItiAl 

Let Si, . . ., S m be the sides of P in cyclic order, and let 
B\, . . ., B m be the interior bisectors of the angles of P, B, 
bisecting the angle between S{ and Si + l if Km and B m 
bisecting that between S m and S\. Then every point of 
Bj has the same distance from Sj as from Sj + i and the 
points of the intersection of consecutive bisectors are at the 
same distance from at least three sides. As these points of 



Fig. 7.14. 

intersection are finite in number there is one or more of them 
which are at a least distance from the sides of the polygon P. 
Suppose that this distance is denoted by X. If we move the 
sides of P inwards by an amount X we obtain a new polygon 
which we denote by Pi. It has fewer sides than P and its 
length L\ and area A\ are related to those of P by the 

A = A x + XU + hX (L - U), i. e. A - A x = J X(L + L x ). 
Thus we have the relation 
(L* - 4 7i A) - (Li» - 4 it At) = (L - U) (L+ U) - 2 nX (L + Lt) 

The parts of the sides of the polygon P which are in excess 
of the length of the parallel sides of P\ can be put together 
to form a polygon circumscribing a circle of radius X. Hence 
L — L\ is larger than the perimeter of this circle, i.e. 2 n/.. 
Thus finally we see that the isoperimetric deficit of P 
exceeds that of Pi. 



The idea underlying this argument can be used to obtain 
an even more precise result. We have seen that the quantity 
L 2 — 4/Ti A is always positive or zero for any convex polygon, 
and is zero for a circle. In fact, it is also always positive 
for any plane non-circular convex set. We may prove this by 
an approximation argument starting from convex polygons. 
Thus we can regard the quantity L 2 — 4-tiA as a measure of 
the degree to which the given convex set differs from a circle, 
the larger this quantity, the less like a circle is the given 
convex set, and the set is a circle if and only if this quantity 
is zero. It is natural to consider whether there are other 
functions which we can define and which have similar 
properties. If there are such functions, it may be possible 
to compare their value with the value given above. There are 
two fairly obvious functions of this nature. One is L — 2 nr, 
where r is the inradius of the convex set. This function 
is always positive or zero, and is positive unless the set is 
a circle. Similar remarks apply to the function 2nR — L, 
where R is the circumradius. Thus we may hope to extend 
our results by comparing the isoperimetric deficit with some 
functions depending upon either L — 2nr or 2tiR — L; in 
fact, the isoperimetric deficit is greater than or equal to 
(I — 2 7ir) 2 and we can prove this by the following extension 
of the previous argument. Let P be a polygon whose sides 
are of total length L and which is of area A (see Fig. 7.15). 
Denote by P<j the polygon whose sides are parallel to those 
of P and at a distance d from those of P measured inwards. 
For small values of d, Pa has the same number of sides as 
has P. But as d increases, the number of sides decreases. 
Denote the length of Pa by La, its area by Ad and its inradius 
by r,i- Also denote by fa the area bounded by the polygon 
whose sides are parallel to those of P<* and which all touch 
a given circle of unit radius. As d increases from zero to r^, 
there are a finite number of values, say d\, . . ., d* at which 
the number of sides of Pd decreases. Suppose that we had 
two values of d, say x and y, both lying between the same 



Area (, 8' 

Perimeter lengTti 2 f, 5 

The polygons p tf all have the same incenlre. 

Fig. 7.15. 

two consecutive d { so that P x and P v have the same number 
of sides. Then it is easy to see that if x is greater than y 
and x — y = d, then we have the following relations: 

A x = A y + L y d + f„d 2 , 

i-'X ~ LJ]/ ~T" 2 f y d, 

r x = r„ + <3, 
U = fy 

By simple algebraic manipulations, we conclude that the 

L x r x — A x — f x r x z 

is identical with the same expression with .v replaced by .»/. 
In other words this quantity is a constant between any two 
consecutive values of the d t . Consider what happens when 
the variable d is allowed to increase through one of the 
values of di. Each of the quantities L dy r dt A d , is a continuous 



function of d. The function fa increases at d, because the 
number of sides of P d decreases. Hence the expression 
given above decreases at di, and thus it decreases as d 
increases from zero to r. As d tends to r, L d , r d and A d 
all decrease to zero and f d maintains a certain value, 
thus f d rd 2 also tends to zero. In other words, the quantity 
which we have above tends to zero, but it is decreasing and 
must therefore always be greater than zero. In particular. 
take d = 0; we deduce that 

Lor — Aq — for 2 > 0, and since fo ^ n, Lq = L, Aq = A, 
we have Lr — A — nr 2 > 0. 

Hence L z — 4-tzA ^ (L — 2 nr) 2 and this is the result which 
we had to prove. This result is also contained in Bonnesen's 

The Greeks never really got to grips with the isoperimetric 
problem. Their difficulties were of a different type from 
those that they encountered in their other famous unsolved 
problems, the duplication of the cube, the trisection of an 
angle and the squaring of a circle. These last three problems 
were unsolved because they were, and are, insoluble. It is 
just impossible to construct a segment of length 2* f *L given 
a segment of length L and using only straight edge and 
compasses, and similarly for the other problems. 

But in the case of the isoperimetric problem the solution 
was there and the Greeks knew that it was, and what it 
was. But the problem involved such a wealth of complex 
concepts that the Greeks were never able to define these 
ideas and to give the problem a satisfactory mathematical 

Today the problem has been generalized out of all 
recognition until finally the name isoperimetric problems 
has been applied by the distinguished mathematicians Polya 
and Szego to problems that are very remote from the original 
one but which, being physical phenomena determined by 


geometric properties, can be tackled by methods similar to 
those of J. Steincr. Some of these can be found in the 
references given below. 

References for further reading 

KoGLESTON, Convexity, Cambridge University Press. 

ECGLESTON, Problems in Euclidean Space, Pergamon Piress. 

LYUSTERNIK, Convex Figures and Polyhedra. 

PrtLYA and SZEGO , Isoperimetric Inequalities in Mathematical 

Physics, Princeton University Press. 

YAGLOM and BOLTYANSKH, Convex Figures, Holt Reinhart & 

The seven lectures included in this 
book formed the programme of the 
1965 Conference in Mathematics 
held at Bedford College, London. 
They are published in this volume 
because they were given by pro- 
fessional mathematicians on 
subjects of current mathematical 
interest, and will be of value to a 
far wider audience than that able to 
attend the Conference. The lectures 
are primarily designed for students 
who are in their last or penultimate 
years at school and who will sub- 
sequently take a mathematics 
degree, orone in which mathematics 
forms a major subject. 

Publication of the lectures in this 
book will also considerably extend 
the effect produced by their 
presentation at the Conference; 
that is, the achieving of an impact 
which greatly helps to create 
interest by introducing students 
to branches of mathematics which 
will be novel to them. An interesting 
point is that two of the lectures 
givea simplified accountof research 
work recently presented by the